Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/mm/page_alloc.c |
| 3 | * |
| 4 | * Manages the free list, the system allocates free pages here. |
| 5 | * Note that kmalloc() lives in slab.c |
| 6 | * |
| 7 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds |
| 8 | * Swap reorganised 29.12.95, Stephen Tweedie |
| 9 | * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999 |
| 10 | * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999 |
| 11 | * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999 |
| 12 | * Zone balancing, Kanoj Sarcar, SGI, Jan 2000 |
| 13 | * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002 |
| 14 | * (lots of bits borrowed from Ingo Molnar & Andrew Morton) |
| 15 | */ |
| 16 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 17 | #include <linux/stddef.h> |
| 18 | #include <linux/mm.h> |
| 19 | #include <linux/swap.h> |
| 20 | #include <linux/interrupt.h> |
| 21 | #include <linux/pagemap.h> |
KOSAKI Motohiro | 10ed273 | 2008-03-04 14:28:32 -0800 | [diff] [blame] | 22 | #include <linux/jiffies.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 23 | #include <linux/bootmem.h> |
Yinghai Lu | edbe7d2 | 2010-08-25 13:39:16 -0700 | [diff] [blame] | 24 | #include <linux/memblock.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 25 | #include <linux/compiler.h> |
Randy Dunlap | 9f15833 | 2005-09-13 01:25:16 -0700 | [diff] [blame] | 26 | #include <linux/kernel.h> |
Vegard Nossum | b1eeab6 | 2008-11-25 16:55:53 +0100 | [diff] [blame] | 27 | #include <linux/kmemcheck.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 28 | #include <linux/module.h> |
| 29 | #include <linux/suspend.h> |
| 30 | #include <linux/pagevec.h> |
| 31 | #include <linux/blkdev.h> |
| 32 | #include <linux/slab.h> |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 33 | #include <linux/ratelimit.h> |
David Rientjes | 5a3135c2 | 2007-10-16 23:25:53 -0700 | [diff] [blame] | 34 | #include <linux/oom.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 35 | #include <linux/notifier.h> |
| 36 | #include <linux/topology.h> |
| 37 | #include <linux/sysctl.h> |
| 38 | #include <linux/cpu.h> |
| 39 | #include <linux/cpuset.h> |
Dave Hansen | bdc8cb9 | 2005-10-29 18:16:53 -0700 | [diff] [blame] | 40 | #include <linux/memory_hotplug.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 41 | #include <linux/nodemask.h> |
| 42 | #include <linux/vmalloc.h> |
KOSAKI Motohiro | a6cccdc | 2011-05-24 17:11:33 -0700 | [diff] [blame] | 43 | #include <linux/vmstat.h> |
Christoph Lameter | 4be38e3 | 2006-01-06 00:11:17 -0800 | [diff] [blame] | 44 | #include <linux/mempolicy.h> |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 45 | #include <linux/stop_machine.h> |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 46 | #include <linux/sort.h> |
| 47 | #include <linux/pfn.h> |
Andrew Morton | 3fcfab1 | 2006-10-19 23:28:16 -0700 | [diff] [blame] | 48 | #include <linux/backing-dev.h> |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 49 | #include <linux/fault-inject.h> |
KAMEZAWA Hiroyuki | a5d76b5 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 50 | #include <linux/page-isolation.h> |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 51 | #include <linux/page_cgroup.h> |
Thomas Gleixner | 3ac7fe5 | 2008-04-30 00:55:01 -0700 | [diff] [blame] | 52 | #include <linux/debugobjects.h> |
Catalin Marinas | dbb1f81 | 2009-06-11 13:23:19 +0100 | [diff] [blame] | 53 | #include <linux/kmemleak.h> |
Robert Jennings | 925cc71 | 2009-12-17 14:44:38 +0000 | [diff] [blame] | 54 | #include <linux/memory.h> |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 55 | #include <linux/compaction.h> |
Mel Gorman | 0d3d062 | 2009-09-21 17:02:44 -0700 | [diff] [blame] | 56 | #include <trace/events/kmem.h> |
Wu Fengguang | 718a382 | 2010-03-10 15:20:43 -0800 | [diff] [blame] | 57 | #include <linux/ftrace_event.h> |
Daisuke Nishimura | f212ad7 | 2011-03-23 16:42:25 -0700 | [diff] [blame] | 58 | #include <linux/memcontrol.h> |
Linus Torvalds | 268bb0c | 2011-05-20 12:50:29 -0700 | [diff] [blame] | 59 | #include <linux/prefetch.h> |
Stanislaw Gruszka | c0a32fc5 | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 60 | #include <linux/page-debug-flags.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 61 | |
| 62 | #include <asm/tlbflush.h> |
Andrew Morton | ac924c6 | 2006-05-15 09:43:59 -0700 | [diff] [blame] | 63 | #include <asm/div64.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 64 | #include "internal.h" |
| 65 | |
Lee Schermerhorn | 7281201 | 2010-05-26 14:44:56 -0700 | [diff] [blame] | 66 | #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID |
| 67 | DEFINE_PER_CPU(int, numa_node); |
| 68 | EXPORT_PER_CPU_SYMBOL(numa_node); |
| 69 | #endif |
| 70 | |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 71 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
| 72 | /* |
| 73 | * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly. |
| 74 | * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined. |
| 75 | * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem() |
| 76 | * defined in <linux/topology.h>. |
| 77 | */ |
| 78 | DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */ |
| 79 | EXPORT_PER_CPU_SYMBOL(_numa_mem_); |
| 80 | #endif |
| 81 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 82 | /* |
Christoph Lameter | 1380891 | 2007-10-16 01:25:27 -0700 | [diff] [blame] | 83 | * Array of node states. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 84 | */ |
Christoph Lameter | 1380891 | 2007-10-16 01:25:27 -0700 | [diff] [blame] | 85 | nodemask_t node_states[NR_NODE_STATES] __read_mostly = { |
| 86 | [N_POSSIBLE] = NODE_MASK_ALL, |
| 87 | [N_ONLINE] = { { [0] = 1UL } }, |
| 88 | #ifndef CONFIG_NUMA |
| 89 | [N_NORMAL_MEMORY] = { { [0] = 1UL } }, |
| 90 | #ifdef CONFIG_HIGHMEM |
| 91 | [N_HIGH_MEMORY] = { { [0] = 1UL } }, |
| 92 | #endif |
| 93 | [N_CPU] = { { [0] = 1UL } }, |
| 94 | #endif /* NUMA */ |
| 95 | }; |
| 96 | EXPORT_SYMBOL(node_states); |
| 97 | |
Ravikiran G Thirumalai | 6c231b7 | 2005-09-06 15:17:45 -0700 | [diff] [blame] | 98 | unsigned long totalram_pages __read_mostly; |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 99 | unsigned long totalreserve_pages __read_mostly; |
Johannes Weiner | ab8fabd | 2012-01-10 15:07:42 -0800 | [diff] [blame] | 100 | /* |
| 101 | * When calculating the number of globally allowed dirty pages, there |
| 102 | * is a certain number of per-zone reserves that should not be |
| 103 | * considered dirtyable memory. This is the sum of those reserves |
| 104 | * over all existing zones that contribute dirtyable memory. |
| 105 | */ |
| 106 | unsigned long dirty_balance_reserve __read_mostly; |
| 107 | |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 108 | int percpu_pagelist_fraction; |
Benjamin Herrenschmidt | dcce284 | 2009-06-18 13:24:12 +1000 | [diff] [blame] | 109 | gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 110 | |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 111 | #ifdef CONFIG_PM_SLEEP |
| 112 | /* |
| 113 | * The following functions are used by the suspend/hibernate code to temporarily |
| 114 | * change gfp_allowed_mask in order to avoid using I/O during memory allocations |
| 115 | * while devices are suspended. To avoid races with the suspend/hibernate code, |
| 116 | * they should always be called with pm_mutex held (gfp_allowed_mask also should |
| 117 | * only be modified with pm_mutex held, unless the suspend/hibernate code is |
| 118 | * guaranteed not to run in parallel with that modification). |
| 119 | */ |
Rafael J. Wysocki | c9e664f | 2010-12-03 22:57:45 +0100 | [diff] [blame] | 120 | |
| 121 | static gfp_t saved_gfp_mask; |
| 122 | |
| 123 | void pm_restore_gfp_mask(void) |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 124 | { |
| 125 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
Rafael J. Wysocki | c9e664f | 2010-12-03 22:57:45 +0100 | [diff] [blame] | 126 | if (saved_gfp_mask) { |
| 127 | gfp_allowed_mask = saved_gfp_mask; |
| 128 | saved_gfp_mask = 0; |
| 129 | } |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 130 | } |
| 131 | |
Rafael J. Wysocki | c9e664f | 2010-12-03 22:57:45 +0100 | [diff] [blame] | 132 | void pm_restrict_gfp_mask(void) |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 133 | { |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 134 | WARN_ON(!mutex_is_locked(&pm_mutex)); |
Rafael J. Wysocki | c9e664f | 2010-12-03 22:57:45 +0100 | [diff] [blame] | 135 | WARN_ON(saved_gfp_mask); |
| 136 | saved_gfp_mask = gfp_allowed_mask; |
| 137 | gfp_allowed_mask &= ~GFP_IOFS; |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 138 | } |
Mel Gorman | f90ac39 | 2012-01-10 15:07:15 -0800 | [diff] [blame] | 139 | |
| 140 | bool pm_suspended_storage(void) |
| 141 | { |
| 142 | if ((gfp_allowed_mask & GFP_IOFS) == GFP_IOFS) |
| 143 | return false; |
| 144 | return true; |
| 145 | } |
Rafael J. Wysocki | 452aa69 | 2010-03-05 13:42:13 -0800 | [diff] [blame] | 146 | #endif /* CONFIG_PM_SLEEP */ |
| 147 | |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 148 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
| 149 | int pageblock_order __read_mostly; |
| 150 | #endif |
| 151 | |
Hugh Dickins | d98c7a0 | 2006-02-14 13:52:59 -0800 | [diff] [blame] | 152 | static void __free_pages_ok(struct page *page, unsigned int order); |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 153 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 154 | /* |
| 155 | * results with 256, 32 in the lowmem_reserve sysctl: |
| 156 | * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high) |
| 157 | * 1G machine -> (16M dma, 784M normal, 224M high) |
| 158 | * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA |
| 159 | * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL |
| 160 | * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA |
Andi Kleen | a2f1b42 | 2005-11-05 17:25:53 +0100 | [diff] [blame] | 161 | * |
| 162 | * TBD: should special case ZONE_DMA32 machines here - in those we normally |
| 163 | * don't need any ZONE_NORMAL reservation |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 164 | */ |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 165 | int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = { |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 166 | #ifdef CONFIG_ZONE_DMA |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 167 | 256, |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 168 | #endif |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 169 | #ifdef CONFIG_ZONE_DMA32 |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 170 | 256, |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 171 | #endif |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 172 | #ifdef CONFIG_HIGHMEM |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 173 | 32, |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 174 | #endif |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 175 | 32, |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 176 | }; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 177 | |
| 178 | EXPORT_SYMBOL(totalram_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 179 | |
Helge Deller | 15ad7cd | 2006-12-06 20:40:36 -0800 | [diff] [blame] | 180 | static char * const zone_names[MAX_NR_ZONES] = { |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 181 | #ifdef CONFIG_ZONE_DMA |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 182 | "DMA", |
Christoph Lameter | 4b51d66 | 2007-02-10 01:43:10 -0800 | [diff] [blame] | 183 | #endif |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 184 | #ifdef CONFIG_ZONE_DMA32 |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 185 | "DMA32", |
Christoph Lameter | fb0e794 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 186 | #endif |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 187 | "Normal", |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 188 | #ifdef CONFIG_HIGHMEM |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 189 | "HighMem", |
Christoph Lameter | e53ef38 | 2006-09-25 23:31:14 -0700 | [diff] [blame] | 190 | #endif |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 191 | "Movable", |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 192 | }; |
| 193 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 194 | int min_free_kbytes = 1024; |
| 195 | |
Jan Beulich | 2c85f51 | 2009-09-21 17:03:07 -0700 | [diff] [blame] | 196 | static unsigned long __meminitdata nr_kernel_pages; |
| 197 | static unsigned long __meminitdata nr_all_pages; |
Yasunori Goto | a3142c8 | 2007-05-08 00:23:07 -0700 | [diff] [blame] | 198 | static unsigned long __meminitdata dma_reserve; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 199 | |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 200 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
| 201 | static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES]; |
| 202 | static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES]; |
| 203 | static unsigned long __initdata required_kernelcore; |
| 204 | static unsigned long __initdata required_movablecore; |
| 205 | static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES]; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 206 | |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 207 | /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */ |
| 208 | int movable_zone; |
| 209 | EXPORT_SYMBOL(movable_zone); |
| 210 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 211 | |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 212 | #if MAX_NUMNODES > 1 |
| 213 | int nr_node_ids __read_mostly = MAX_NUMNODES; |
Christoph Lameter | 62bc62a | 2009-06-16 15:32:15 -0700 | [diff] [blame] | 214 | int nr_online_nodes __read_mostly = 1; |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 215 | EXPORT_SYMBOL(nr_node_ids); |
Christoph Lameter | 62bc62a | 2009-06-16 15:32:15 -0700 | [diff] [blame] | 216 | EXPORT_SYMBOL(nr_online_nodes); |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 217 | #endif |
| 218 | |
Mel Gorman | 9ef9acb | 2007-10-16 01:25:54 -0700 | [diff] [blame] | 219 | int page_group_by_mobility_disabled __read_mostly; |
| 220 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 221 | static void set_pageblock_migratetype(struct page *page, int migratetype) |
| 222 | { |
Mel Gorman | 49255c6 | 2009-06-16 15:31:58 -0700 | [diff] [blame] | 223 | |
| 224 | if (unlikely(page_group_by_mobility_disabled)) |
| 225 | migratetype = MIGRATE_UNMOVABLE; |
| 226 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 227 | set_pageblock_flags_group(page, (unsigned long)migratetype, |
| 228 | PB_migrate, PB_migrate_end); |
| 229 | } |
| 230 | |
Rafael J. Wysocki | 7f33d49 | 2009-06-16 15:32:41 -0700 | [diff] [blame] | 231 | bool oom_killer_disabled __read_mostly; |
| 232 | |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 233 | #ifdef CONFIG_DEBUG_VM |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 234 | static int page_outside_zone_boundaries(struct zone *zone, struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 235 | { |
Dave Hansen | bdc8cb9 | 2005-10-29 18:16:53 -0700 | [diff] [blame] | 236 | int ret = 0; |
| 237 | unsigned seq; |
| 238 | unsigned long pfn = page_to_pfn(page); |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 239 | |
Dave Hansen | bdc8cb9 | 2005-10-29 18:16:53 -0700 | [diff] [blame] | 240 | do { |
| 241 | seq = zone_span_seqbegin(zone); |
| 242 | if (pfn >= zone->zone_start_pfn + zone->spanned_pages) |
| 243 | ret = 1; |
| 244 | else if (pfn < zone->zone_start_pfn) |
| 245 | ret = 1; |
| 246 | } while (zone_span_seqretry(zone, seq)); |
| 247 | |
| 248 | return ret; |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 249 | } |
| 250 | |
| 251 | static int page_is_consistent(struct zone *zone, struct page *page) |
| 252 | { |
Andy Whitcroft | 14e0729 | 2007-05-06 14:49:14 -0700 | [diff] [blame] | 253 | if (!pfn_valid_within(page_to_pfn(page))) |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 254 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 255 | if (zone != page_zone(page)) |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 256 | return 0; |
| 257 | |
| 258 | return 1; |
| 259 | } |
| 260 | /* |
| 261 | * Temporary debugging check for pages not lying within a given zone. |
| 262 | */ |
| 263 | static int bad_range(struct zone *zone, struct page *page) |
| 264 | { |
| 265 | if (page_outside_zone_boundaries(zone, page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 266 | return 1; |
Dave Hansen | c6a57e1 | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 267 | if (!page_is_consistent(zone, page)) |
| 268 | return 1; |
| 269 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 270 | return 0; |
| 271 | } |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 272 | #else |
| 273 | static inline int bad_range(struct zone *zone, struct page *page) |
| 274 | { |
| 275 | return 0; |
| 276 | } |
| 277 | #endif |
| 278 | |
Nick Piggin | 224abf9 | 2006-01-06 00:11:11 -0800 | [diff] [blame] | 279 | static void bad_page(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 280 | { |
Hugh Dickins | d936cf9 | 2009-01-06 14:40:12 -0800 | [diff] [blame] | 281 | static unsigned long resume; |
| 282 | static unsigned long nr_shown; |
| 283 | static unsigned long nr_unshown; |
| 284 | |
Wu Fengguang | 2a7684a | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 285 | /* Don't complain about poisoned pages */ |
| 286 | if (PageHWPoison(page)) { |
Andrea Arcangeli | ef2b4b9 | 2011-03-18 00:16:35 +0100 | [diff] [blame] | 287 | reset_page_mapcount(page); /* remove PageBuddy */ |
Wu Fengguang | 2a7684a | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 288 | return; |
| 289 | } |
| 290 | |
Hugh Dickins | d936cf9 | 2009-01-06 14:40:12 -0800 | [diff] [blame] | 291 | /* |
| 292 | * Allow a burst of 60 reports, then keep quiet for that minute; |
| 293 | * or allow a steady drip of one report per second. |
| 294 | */ |
| 295 | if (nr_shown == 60) { |
| 296 | if (time_before(jiffies, resume)) { |
| 297 | nr_unshown++; |
| 298 | goto out; |
| 299 | } |
| 300 | if (nr_unshown) { |
Hugh Dickins | 1e9e636 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 301 | printk(KERN_ALERT |
| 302 | "BUG: Bad page state: %lu messages suppressed\n", |
Hugh Dickins | d936cf9 | 2009-01-06 14:40:12 -0800 | [diff] [blame] | 303 | nr_unshown); |
| 304 | nr_unshown = 0; |
| 305 | } |
| 306 | nr_shown = 0; |
| 307 | } |
| 308 | if (nr_shown++ == 0) |
| 309 | resume = jiffies + 60 * HZ; |
| 310 | |
Hugh Dickins | 1e9e636 | 2009-01-06 14:40:13 -0800 | [diff] [blame] | 311 | printk(KERN_ALERT "BUG: Bad page state in process %s pfn:%05lx\n", |
Hugh Dickins | 3dc1474 | 2009-01-06 14:40:08 -0800 | [diff] [blame] | 312 | current->comm, page_to_pfn(page)); |
Wu Fengguang | 718a382 | 2010-03-10 15:20:43 -0800 | [diff] [blame] | 313 | dump_page(page); |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 314 | |
Dave Jones | 4f31888 | 2011-10-31 17:07:24 -0700 | [diff] [blame] | 315 | print_modules(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 316 | dump_stack(); |
Hugh Dickins | d936cf9 | 2009-01-06 14:40:12 -0800 | [diff] [blame] | 317 | out: |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 318 | /* Leave bad fields for debug, except PageBuddy could make trouble */ |
Andrea Arcangeli | ef2b4b9 | 2011-03-18 00:16:35 +0100 | [diff] [blame] | 319 | reset_page_mapcount(page); /* remove PageBuddy */ |
Randy Dunlap | 9f15833 | 2005-09-13 01:25:16 -0700 | [diff] [blame] | 320 | add_taint(TAINT_BAD_PAGE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 321 | } |
| 322 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 323 | /* |
| 324 | * Higher-order pages are called "compound pages". They are structured thusly: |
| 325 | * |
| 326 | * The first PAGE_SIZE page is called the "head page". |
| 327 | * |
| 328 | * The remaining PAGE_SIZE pages are called "tail pages". |
| 329 | * |
Wang Sheng-Hui | 6416b9f | 2011-11-17 10:53:50 +0100 | [diff] [blame] | 330 | * All pages have PG_compound set. All tail pages have their ->first_page |
| 331 | * pointing at the head page. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 332 | * |
Hugh Dickins | 41d78ba | 2006-02-14 13:52:58 -0800 | [diff] [blame] | 333 | * The first tail page's ->lru.next holds the address of the compound page's |
| 334 | * put_page() function. Its ->lru.prev holds the order of allocation. |
| 335 | * This usage means that zero-order pages may not be compound. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 336 | */ |
Hugh Dickins | d98c7a0 | 2006-02-14 13:52:59 -0800 | [diff] [blame] | 337 | |
| 338 | static void free_compound_page(struct page *page) |
| 339 | { |
Christoph Lameter | d85f338 | 2007-05-06 14:49:39 -0700 | [diff] [blame] | 340 | __free_pages_ok(page, compound_order(page)); |
Hugh Dickins | d98c7a0 | 2006-02-14 13:52:59 -0800 | [diff] [blame] | 341 | } |
| 342 | |
Andi Kleen | 01ad1c0 | 2008-07-23 21:27:46 -0700 | [diff] [blame] | 343 | void prep_compound_page(struct page *page, unsigned long order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 344 | { |
| 345 | int i; |
| 346 | int nr_pages = 1 << order; |
| 347 | |
Andy Whitcroft | 33f2ef8 | 2006-12-06 20:33:32 -0800 | [diff] [blame] | 348 | set_compound_page_dtor(page, free_compound_page); |
Christoph Lameter | d85f338 | 2007-05-06 14:49:39 -0700 | [diff] [blame] | 349 | set_compound_order(page, order); |
Christoph Lameter | 6d77795 | 2007-05-06 14:49:40 -0700 | [diff] [blame] | 350 | __SetPageHead(page); |
Andy Whitcroft | 18229df | 2008-11-06 12:53:27 -0800 | [diff] [blame] | 351 | for (i = 1; i < nr_pages; i++) { |
| 352 | struct page *p = page + i; |
Christoph Lameter | d85f338 | 2007-05-06 14:49:39 -0700 | [diff] [blame] | 353 | __SetPageTail(p); |
Youquan Song | 58a84aa | 2011-12-08 14:34:18 -0800 | [diff] [blame] | 354 | set_page_count(p, 0); |
Christoph Lameter | d85f338 | 2007-05-06 14:49:39 -0700 | [diff] [blame] | 355 | p->first_page = page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 356 | } |
| 357 | } |
| 358 | |
Andrea Arcangeli | 59ff421 | 2011-01-13 15:46:38 -0800 | [diff] [blame] | 359 | /* update __split_huge_page_refcount if you change this function */ |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 360 | static int destroy_compound_page(struct page *page, unsigned long order) |
Andy Whitcroft | 18229df | 2008-11-06 12:53:27 -0800 | [diff] [blame] | 361 | { |
| 362 | int i; |
| 363 | int nr_pages = 1 << order; |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 364 | int bad = 0; |
Andy Whitcroft | 18229df | 2008-11-06 12:53:27 -0800 | [diff] [blame] | 365 | |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 366 | if (unlikely(compound_order(page) != order) || |
| 367 | unlikely(!PageHead(page))) { |
Nick Piggin | 224abf9 | 2006-01-06 00:11:11 -0800 | [diff] [blame] | 368 | bad_page(page); |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 369 | bad++; |
| 370 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 371 | |
Christoph Lameter | 6d77795 | 2007-05-06 14:49:40 -0700 | [diff] [blame] | 372 | __ClearPageHead(page); |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 373 | |
Andy Whitcroft | 18229df | 2008-11-06 12:53:27 -0800 | [diff] [blame] | 374 | for (i = 1; i < nr_pages; i++) { |
| 375 | struct page *p = page + i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 376 | |
Alexey Zaytsev | e713a21 | 2009-01-10 02:47:57 +0300 | [diff] [blame] | 377 | if (unlikely(!PageTail(p) || (p->first_page != page))) { |
Nick Piggin | 224abf9 | 2006-01-06 00:11:11 -0800 | [diff] [blame] | 378 | bad_page(page); |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 379 | bad++; |
| 380 | } |
Christoph Lameter | d85f338 | 2007-05-06 14:49:39 -0700 | [diff] [blame] | 381 | __ClearPageTail(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 382 | } |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 383 | |
| 384 | return bad; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 385 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 386 | |
Nick Piggin | 17cf440 | 2006-03-22 00:08:41 -0800 | [diff] [blame] | 387 | static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags) |
| 388 | { |
| 389 | int i; |
| 390 | |
Andrew Morton | 6626c5d | 2006-03-22 00:08:42 -0800 | [diff] [blame] | 391 | /* |
| 392 | * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO |
| 393 | * and __GFP_HIGHMEM from hard or soft interrupt context. |
| 394 | */ |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 395 | VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt()); |
Nick Piggin | 17cf440 | 2006-03-22 00:08:41 -0800 | [diff] [blame] | 396 | for (i = 0; i < (1 << order); i++) |
| 397 | clear_highpage(page + i); |
| 398 | } |
| 399 | |
Stanislaw Gruszka | c0a32fc5 | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 400 | #ifdef CONFIG_DEBUG_PAGEALLOC |
| 401 | unsigned int _debug_guardpage_minorder; |
| 402 | |
| 403 | static int __init debug_guardpage_minorder_setup(char *buf) |
| 404 | { |
| 405 | unsigned long res; |
| 406 | |
| 407 | if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) { |
| 408 | printk(KERN_ERR "Bad debug_guardpage_minorder value\n"); |
| 409 | return 0; |
| 410 | } |
| 411 | _debug_guardpage_minorder = res; |
| 412 | printk(KERN_INFO "Setting debug_guardpage_minorder to %lu\n", res); |
| 413 | return 0; |
| 414 | } |
| 415 | __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup); |
| 416 | |
| 417 | static inline void set_page_guard_flag(struct page *page) |
| 418 | { |
| 419 | __set_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags); |
| 420 | } |
| 421 | |
| 422 | static inline void clear_page_guard_flag(struct page *page) |
| 423 | { |
| 424 | __clear_bit(PAGE_DEBUG_FLAG_GUARD, &page->debug_flags); |
| 425 | } |
| 426 | #else |
| 427 | static inline void set_page_guard_flag(struct page *page) { } |
| 428 | static inline void clear_page_guard_flag(struct page *page) { } |
| 429 | #endif |
| 430 | |
Andrew Morton | 6aa3001b2 | 2006-04-18 22:20:52 -0700 | [diff] [blame] | 431 | static inline void set_page_order(struct page *page, int order) |
| 432 | { |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 433 | set_page_private(page, order); |
Nick Piggin | 676165a | 2006-04-10 11:21:48 +1000 | [diff] [blame] | 434 | __SetPageBuddy(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 435 | } |
| 436 | |
| 437 | static inline void rmv_page_order(struct page *page) |
| 438 | { |
Nick Piggin | 676165a | 2006-04-10 11:21:48 +1000 | [diff] [blame] | 439 | __ClearPageBuddy(page); |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 440 | set_page_private(page, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 441 | } |
| 442 | |
| 443 | /* |
| 444 | * Locate the struct page for both the matching buddy in our |
| 445 | * pair (buddy1) and the combined O(n+1) page they form (page). |
| 446 | * |
| 447 | * 1) Any buddy B1 will have an order O twin B2 which satisfies |
| 448 | * the following equation: |
| 449 | * B2 = B1 ^ (1 << O) |
| 450 | * For example, if the starting buddy (buddy2) is #8 its order |
| 451 | * 1 buddy is #10: |
| 452 | * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10 |
| 453 | * |
| 454 | * 2) Any buddy B will have an order O+1 parent P which |
| 455 | * satisfies the following equation: |
| 456 | * P = B & ~(1 << O) |
| 457 | * |
Andreas Mohr | d6e05ed | 2006-06-26 18:35:02 +0200 | [diff] [blame] | 458 | * Assumption: *_mem_map is contiguous at least up to MAX_ORDER |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 459 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 460 | static inline unsigned long |
KyongHo Cho | 43506fa | 2011-01-13 15:47:24 -0800 | [diff] [blame] | 461 | __find_buddy_index(unsigned long page_idx, unsigned int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 462 | { |
KyongHo Cho | 43506fa | 2011-01-13 15:47:24 -0800 | [diff] [blame] | 463 | return page_idx ^ (1 << order); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 464 | } |
| 465 | |
| 466 | /* |
| 467 | * This function checks whether a page is free && is the buddy |
| 468 | * we can do coalesce a page and its buddy if |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 469 | * (a) the buddy is not in a hole && |
Nick Piggin | 676165a | 2006-04-10 11:21:48 +1000 | [diff] [blame] | 470 | * (b) the buddy is in the buddy system && |
Andy Whitcroft | cb2b95e | 2006-06-23 02:03:01 -0700 | [diff] [blame] | 471 | * (c) a page and its buddy have the same order && |
| 472 | * (d) a page and its buddy are in the same zone. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 473 | * |
Andrea Arcangeli | 5f24ce5 | 2011-01-13 15:47:00 -0800 | [diff] [blame] | 474 | * For recording whether a page is in the buddy system, we set ->_mapcount -2. |
| 475 | * Setting, clearing, and testing _mapcount -2 is serialized by zone->lock. |
Nick Piggin | 676165a | 2006-04-10 11:21:48 +1000 | [diff] [blame] | 476 | * |
| 477 | * For recording page's order, we use page_private(page). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 478 | */ |
Andy Whitcroft | cb2b95e | 2006-06-23 02:03:01 -0700 | [diff] [blame] | 479 | static inline int page_is_buddy(struct page *page, struct page *buddy, |
| 480 | int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 481 | { |
Andy Whitcroft | 14e0729 | 2007-05-06 14:49:14 -0700 | [diff] [blame] | 482 | if (!pfn_valid_within(page_to_pfn(buddy))) |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 483 | return 0; |
Nick Piggin | 13e7444 | 2006-01-06 00:10:58 -0800 | [diff] [blame] | 484 | |
Andy Whitcroft | cb2b95e | 2006-06-23 02:03:01 -0700 | [diff] [blame] | 485 | if (page_zone_id(page) != page_zone_id(buddy)) |
| 486 | return 0; |
| 487 | |
Stanislaw Gruszka | c0a32fc5 | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 488 | if (page_is_guard(buddy) && page_order(buddy) == order) { |
| 489 | VM_BUG_ON(page_count(buddy) != 0); |
| 490 | return 1; |
| 491 | } |
| 492 | |
Andy Whitcroft | cb2b95e | 2006-06-23 02:03:01 -0700 | [diff] [blame] | 493 | if (PageBuddy(buddy) && page_order(buddy) == order) { |
Nick Piggin | a3af9c3 | 2009-06-16 15:32:10 -0700 | [diff] [blame] | 494 | VM_BUG_ON(page_count(buddy) != 0); |
Andrew Morton | 6aa3001b2 | 2006-04-18 22:20:52 -0700 | [diff] [blame] | 495 | return 1; |
Nick Piggin | 676165a | 2006-04-10 11:21:48 +1000 | [diff] [blame] | 496 | } |
Andrew Morton | 6aa3001b2 | 2006-04-18 22:20:52 -0700 | [diff] [blame] | 497 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 498 | } |
| 499 | |
| 500 | /* |
| 501 | * Freeing function for a buddy system allocator. |
| 502 | * |
| 503 | * The concept of a buddy system is to maintain direct-mapped table |
| 504 | * (containing bit values) for memory blocks of various "orders". |
| 505 | * The bottom level table contains the map for the smallest allocatable |
| 506 | * units of memory (here, pages), and each level above it describes |
| 507 | * pairs of units from the levels below, hence, "buddies". |
| 508 | * At a high level, all that happens here is marking the table entry |
| 509 | * at the bottom level available, and propagating the changes upward |
| 510 | * as necessary, plus some accounting needed to play nicely with other |
| 511 | * parts of the VM system. |
| 512 | * At each level, we keep a list of pages, which are heads of continuous |
Andrea Arcangeli | 5f24ce5 | 2011-01-13 15:47:00 -0800 | [diff] [blame] | 513 | * free pages of length of (1 << order) and marked with _mapcount -2. Page's |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 514 | * order is recorded in page_private(page) field. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 515 | * So when we are allocating or freeing one, we can derive the state of the |
| 516 | * other. That is, if we allocate a small block, and both were |
| 517 | * free, the remainder of the region must be split into blocks. |
| 518 | * If a block is freed, and its buddy is also free, then this |
| 519 | * triggers coalescing into a block of larger size. |
| 520 | * |
| 521 | * -- wli |
| 522 | */ |
| 523 | |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 524 | static inline void __free_one_page(struct page *page, |
Mel Gorman | ed0ae21 | 2009-06-16 15:32:07 -0700 | [diff] [blame] | 525 | struct zone *zone, unsigned int order, |
| 526 | int migratetype) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 527 | { |
| 528 | unsigned long page_idx; |
Corrado Zoccolo | 6dda9d5 | 2010-05-24 14:31:54 -0700 | [diff] [blame] | 529 | unsigned long combined_idx; |
KyongHo Cho | 43506fa | 2011-01-13 15:47:24 -0800 | [diff] [blame] | 530 | unsigned long uninitialized_var(buddy_idx); |
Corrado Zoccolo | 6dda9d5 | 2010-05-24 14:31:54 -0700 | [diff] [blame] | 531 | struct page *buddy; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 532 | |
Nick Piggin | 224abf9 | 2006-01-06 00:11:11 -0800 | [diff] [blame] | 533 | if (unlikely(PageCompound(page))) |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 534 | if (unlikely(destroy_compound_page(page, order))) |
| 535 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 536 | |
Mel Gorman | ed0ae21 | 2009-06-16 15:32:07 -0700 | [diff] [blame] | 537 | VM_BUG_ON(migratetype == -1); |
| 538 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 539 | page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1); |
| 540 | |
Mel Gorman | f2260e6 | 2009-06-16 15:32:13 -0700 | [diff] [blame] | 541 | VM_BUG_ON(page_idx & ((1 << order) - 1)); |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 542 | VM_BUG_ON(bad_range(zone, page)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 543 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 544 | while (order < MAX_ORDER-1) { |
KyongHo Cho | 43506fa | 2011-01-13 15:47:24 -0800 | [diff] [blame] | 545 | buddy_idx = __find_buddy_index(page_idx, order); |
| 546 | buddy = page + (buddy_idx - page_idx); |
Andy Whitcroft | cb2b95e | 2006-06-23 02:03:01 -0700 | [diff] [blame] | 547 | if (!page_is_buddy(page, buddy, order)) |
Andy Whitcroft | 3c82d0c | 2008-07-23 21:27:11 -0700 | [diff] [blame] | 548 | break; |
Stanislaw Gruszka | c0a32fc5 | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 549 | /* |
| 550 | * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page, |
| 551 | * merge with it and move up one order. |
| 552 | */ |
| 553 | if (page_is_guard(buddy)) { |
| 554 | clear_page_guard_flag(buddy); |
| 555 | set_page_private(page, 0); |
| 556 | __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order); |
| 557 | } else { |
| 558 | list_del(&buddy->lru); |
| 559 | zone->free_area[order].nr_free--; |
| 560 | rmv_page_order(buddy); |
| 561 | } |
KyongHo Cho | 43506fa | 2011-01-13 15:47:24 -0800 | [diff] [blame] | 562 | combined_idx = buddy_idx & page_idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 563 | page = page + (combined_idx - page_idx); |
| 564 | page_idx = combined_idx; |
| 565 | order++; |
| 566 | } |
| 567 | set_page_order(page, order); |
Corrado Zoccolo | 6dda9d5 | 2010-05-24 14:31:54 -0700 | [diff] [blame] | 568 | |
| 569 | /* |
| 570 | * If this is not the largest possible page, check if the buddy |
| 571 | * of the next-highest order is free. If it is, it's possible |
| 572 | * that pages are being freed that will coalesce soon. In case, |
| 573 | * that is happening, add the free page to the tail of the list |
| 574 | * so it's less likely to be used soon and more likely to be merged |
| 575 | * as a higher order page |
| 576 | */ |
Mel Gorman | b7f50cf | 2010-10-26 14:21:11 -0700 | [diff] [blame] | 577 | if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) { |
Corrado Zoccolo | 6dda9d5 | 2010-05-24 14:31:54 -0700 | [diff] [blame] | 578 | struct page *higher_page, *higher_buddy; |
KyongHo Cho | 43506fa | 2011-01-13 15:47:24 -0800 | [diff] [blame] | 579 | combined_idx = buddy_idx & page_idx; |
| 580 | higher_page = page + (combined_idx - page_idx); |
| 581 | buddy_idx = __find_buddy_index(combined_idx, order + 1); |
| 582 | higher_buddy = page + (buddy_idx - combined_idx); |
Corrado Zoccolo | 6dda9d5 | 2010-05-24 14:31:54 -0700 | [diff] [blame] | 583 | if (page_is_buddy(higher_page, higher_buddy, order + 1)) { |
| 584 | list_add_tail(&page->lru, |
| 585 | &zone->free_area[order].free_list[migratetype]); |
| 586 | goto out; |
| 587 | } |
| 588 | } |
| 589 | |
| 590 | list_add(&page->lru, &zone->free_area[order].free_list[migratetype]); |
| 591 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 592 | zone->free_area[order].nr_free++; |
| 593 | } |
| 594 | |
KOSAKI Motohiro | 092cead | 2009-06-16 15:32:17 -0700 | [diff] [blame] | 595 | /* |
| 596 | * free_page_mlock() -- clean up attempts to free and mlocked() page. |
| 597 | * Page should not be on lru, so no need to fix that up. |
| 598 | * free_pages_check() will verify... |
| 599 | */ |
| 600 | static inline void free_page_mlock(struct page *page) |
| 601 | { |
KOSAKI Motohiro | 092cead | 2009-06-16 15:32:17 -0700 | [diff] [blame] | 602 | __dec_zone_page_state(page, NR_MLOCK); |
| 603 | __count_vm_event(UNEVICTABLE_MLOCKFREED); |
| 604 | } |
KOSAKI Motohiro | 092cead | 2009-06-16 15:32:17 -0700 | [diff] [blame] | 605 | |
Nick Piggin | 224abf9 | 2006-01-06 00:11:11 -0800 | [diff] [blame] | 606 | static inline int free_pages_check(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 607 | { |
Nick Piggin | 92be2e33 | 2006-01-06 00:10:57 -0800 | [diff] [blame] | 608 | if (unlikely(page_mapcount(page) | |
| 609 | (page->mapping != NULL) | |
Nick Piggin | a3af9c3 | 2009-06-16 15:32:10 -0700 | [diff] [blame] | 610 | (atomic_read(&page->_count) != 0) | |
Daisuke Nishimura | f212ad7 | 2011-03-23 16:42:25 -0700 | [diff] [blame] | 611 | (page->flags & PAGE_FLAGS_CHECK_AT_FREE) | |
| 612 | (mem_cgroup_bad_page_check(page)))) { |
Nick Piggin | 224abf9 | 2006-01-06 00:11:11 -0800 | [diff] [blame] | 613 | bad_page(page); |
Hugh Dickins | 79f4b7b | 2009-01-06 14:40:05 -0800 | [diff] [blame] | 614 | return 1; |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 615 | } |
Hugh Dickins | 79f4b7b | 2009-01-06 14:40:05 -0800 | [diff] [blame] | 616 | if (page->flags & PAGE_FLAGS_CHECK_AT_PREP) |
| 617 | page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; |
| 618 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 619 | } |
| 620 | |
| 621 | /* |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 622 | * Frees a number of pages from the PCP lists |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 623 | * Assumes all pages on list are in same zone, and of same order. |
Renaud Lienhart | 207f36e | 2005-09-10 00:26:59 -0700 | [diff] [blame] | 624 | * count is the number of pages to free. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 625 | * |
| 626 | * If the zone was previously in an "all pages pinned" state then look to |
| 627 | * see if this freeing clears that state. |
| 628 | * |
| 629 | * And clear the zone's pages_scanned counter, to hold off the "all pages are |
| 630 | * pinned" detection logic. |
| 631 | */ |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 632 | static void free_pcppages_bulk(struct zone *zone, int count, |
| 633 | struct per_cpu_pages *pcp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 634 | { |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 635 | int migratetype = 0; |
Mel Gorman | a6f9edd | 2009-09-21 17:03:20 -0700 | [diff] [blame] | 636 | int batch_free = 0; |
Mel Gorman | 72853e2 | 2010-09-09 16:38:16 -0700 | [diff] [blame] | 637 | int to_free = count; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 638 | |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 639 | spin_lock(&zone->lock); |
KOSAKI Motohiro | 93e4a89 | 2010-03-05 13:41:55 -0800 | [diff] [blame] | 640 | zone->all_unreclaimable = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 641 | zone->pages_scanned = 0; |
Mel Gorman | f2260e6 | 2009-06-16 15:32:13 -0700 | [diff] [blame] | 642 | |
Mel Gorman | 72853e2 | 2010-09-09 16:38:16 -0700 | [diff] [blame] | 643 | while (to_free) { |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 644 | struct page *page; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 645 | struct list_head *list; |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 646 | |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 647 | /* |
Mel Gorman | a6f9edd | 2009-09-21 17:03:20 -0700 | [diff] [blame] | 648 | * Remove pages from lists in a round-robin fashion. A |
| 649 | * batch_free count is maintained that is incremented when an |
| 650 | * empty list is encountered. This is so more pages are freed |
| 651 | * off fuller lists instead of spinning excessively around empty |
| 652 | * lists |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 653 | */ |
| 654 | do { |
Mel Gorman | a6f9edd | 2009-09-21 17:03:20 -0700 | [diff] [blame] | 655 | batch_free++; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 656 | if (++migratetype == MIGRATE_PCPTYPES) |
| 657 | migratetype = 0; |
| 658 | list = &pcp->lists[migratetype]; |
| 659 | } while (list_empty(list)); |
| 660 | |
Namhyung Kim | 1d16871 | 2011-03-22 16:32:45 -0700 | [diff] [blame] | 661 | /* This is the only non-empty list. Free them all. */ |
| 662 | if (batch_free == MIGRATE_PCPTYPES) |
| 663 | batch_free = to_free; |
| 664 | |
Mel Gorman | a6f9edd | 2009-09-21 17:03:20 -0700 | [diff] [blame] | 665 | do { |
| 666 | page = list_entry(list->prev, struct page, lru); |
| 667 | /* must delete as __free_one_page list manipulates */ |
| 668 | list_del(&page->lru); |
Hugh Dickins | a701623 | 2010-01-29 17:46:34 +0000 | [diff] [blame] | 669 | /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */ |
| 670 | __free_one_page(page, zone, 0, page_private(page)); |
| 671 | trace_mm_page_pcpu_drain(page, 0, page_private(page)); |
Mel Gorman | 72853e2 | 2010-09-09 16:38:16 -0700 | [diff] [blame] | 672 | } while (--to_free && --batch_free && !list_empty(list)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 673 | } |
Mel Gorman | 72853e2 | 2010-09-09 16:38:16 -0700 | [diff] [blame] | 674 | __mod_zone_page_state(zone, NR_FREE_PAGES, count); |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 675 | spin_unlock(&zone->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 676 | } |
| 677 | |
Mel Gorman | ed0ae21 | 2009-06-16 15:32:07 -0700 | [diff] [blame] | 678 | static void free_one_page(struct zone *zone, struct page *page, int order, |
| 679 | int migratetype) |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 680 | { |
Christoph Lameter | 006d22d | 2006-09-25 23:31:48 -0700 | [diff] [blame] | 681 | spin_lock(&zone->lock); |
KOSAKI Motohiro | 93e4a89 | 2010-03-05 13:41:55 -0800 | [diff] [blame] | 682 | zone->all_unreclaimable = 0; |
Christoph Lameter | 006d22d | 2006-09-25 23:31:48 -0700 | [diff] [blame] | 683 | zone->pages_scanned = 0; |
Mel Gorman | f2260e6 | 2009-06-16 15:32:13 -0700 | [diff] [blame] | 684 | |
Mel Gorman | ed0ae21 | 2009-06-16 15:32:07 -0700 | [diff] [blame] | 685 | __free_one_page(page, zone, order, migratetype); |
Mel Gorman | 72853e2 | 2010-09-09 16:38:16 -0700 | [diff] [blame] | 686 | __mod_zone_page_state(zone, NR_FREE_PAGES, 1 << order); |
Christoph Lameter | 006d22d | 2006-09-25 23:31:48 -0700 | [diff] [blame] | 687 | spin_unlock(&zone->lock); |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 688 | } |
| 689 | |
KOSAKI Motohiro | ec95f53 | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 690 | static bool free_pages_prepare(struct page *page, unsigned int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 691 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 692 | int i; |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 693 | int bad = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 694 | |
Konstantin Khlebnikov | b413d48 | 2012-01-10 15:07:09 -0800 | [diff] [blame] | 695 | trace_mm_page_free(page, order); |
Vegard Nossum | b1eeab6 | 2008-11-25 16:55:53 +0100 | [diff] [blame] | 696 | kmemcheck_free_shadow(page, order); |
| 697 | |
Andrea Arcangeli | 8dd60a3 | 2011-01-13 15:46:34 -0800 | [diff] [blame] | 698 | if (PageAnon(page)) |
| 699 | page->mapping = NULL; |
| 700 | for (i = 0; i < (1 << order); i++) |
| 701 | bad += free_pages_check(page + i); |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 702 | if (bad) |
KOSAKI Motohiro | ec95f53 | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 703 | return false; |
Hugh Dickins | 689bceb | 2005-11-21 21:32:20 -0800 | [diff] [blame] | 704 | |
Thomas Gleixner | 3ac7fe5 | 2008-04-30 00:55:01 -0700 | [diff] [blame] | 705 | if (!PageHighMem(page)) { |
Nick Piggin | 9858db5 | 2006-10-11 01:21:30 -0700 | [diff] [blame] | 706 | debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order); |
Thomas Gleixner | 3ac7fe5 | 2008-04-30 00:55:01 -0700 | [diff] [blame] | 707 | debug_check_no_obj_freed(page_address(page), |
| 708 | PAGE_SIZE << order); |
| 709 | } |
Nick Piggin | dafb136 | 2006-10-11 01:21:30 -0700 | [diff] [blame] | 710 | arch_free_page(page, order); |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 711 | kernel_map_pages(page, 1 << order, 0); |
Nick Piggin | dafb136 | 2006-10-11 01:21:30 -0700 | [diff] [blame] | 712 | |
KOSAKI Motohiro | ec95f53 | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 713 | return true; |
| 714 | } |
| 715 | |
| 716 | static void __free_pages_ok(struct page *page, unsigned int order) |
| 717 | { |
| 718 | unsigned long flags; |
| 719 | int wasMlocked = __TestClearPageMlocked(page); |
| 720 | |
| 721 | if (!free_pages_prepare(page, order)) |
| 722 | return; |
| 723 | |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 724 | local_irq_save(flags); |
Johannes Weiner | c277331 | 2009-06-19 19:30:56 +0200 | [diff] [blame] | 725 | if (unlikely(wasMlocked)) |
Mel Gorman | da456f1 | 2009-06-16 15:32:08 -0700 | [diff] [blame] | 726 | free_page_mlock(page); |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 727 | __count_vm_events(PGFREE, 1 << order); |
Mel Gorman | ed0ae21 | 2009-06-16 15:32:07 -0700 | [diff] [blame] | 728 | free_one_page(page_zone(page), page, order, |
| 729 | get_pageblock_migratetype(page)); |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 730 | local_irq_restore(flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 731 | } |
| 732 | |
Yasunori Goto | af370fb | 2008-07-23 21:28:17 -0700 | [diff] [blame] | 733 | void __meminit __free_pages_bootmem(struct page *page, unsigned int order) |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 734 | { |
Johannes Weiner | c399307 | 2012-01-10 15:08:10 -0800 | [diff] [blame] | 735 | unsigned int nr_pages = 1 << order; |
| 736 | unsigned int loop; |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 737 | |
Johannes Weiner | c399307 | 2012-01-10 15:08:10 -0800 | [diff] [blame] | 738 | prefetchw(page); |
| 739 | for (loop = 0; loop < nr_pages; loop++) { |
| 740 | struct page *p = &page[loop]; |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 741 | |
Johannes Weiner | c399307 | 2012-01-10 15:08:10 -0800 | [diff] [blame] | 742 | if (loop + 1 < nr_pages) |
| 743 | prefetchw(p + 1); |
| 744 | __ClearPageReserved(p); |
| 745 | set_page_count(p, 0); |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 746 | } |
Johannes Weiner | c399307 | 2012-01-10 15:08:10 -0800 | [diff] [blame] | 747 | |
| 748 | set_page_refcounted(page); |
| 749 | __free_pages(page, order); |
David Howells | a226f6c | 2006-01-06 00:11:08 -0800 | [diff] [blame] | 750 | } |
| 751 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 752 | |
| 753 | /* |
| 754 | * The order of subdivision here is critical for the IO subsystem. |
| 755 | * Please do not alter this order without good reasons and regression |
| 756 | * testing. Specifically, as large blocks of memory are subdivided, |
| 757 | * the order in which smaller blocks are delivered depends on the order |
| 758 | * they're subdivided in this function. This is the primary factor |
| 759 | * influencing the order in which pages are delivered to the IO |
| 760 | * subsystem according to empirical testing, and this is also justified |
| 761 | * by considering the behavior of a buddy system containing a single |
| 762 | * large block of memory acted on by a series of small allocations. |
| 763 | * This behavior is a critical factor in sglist merging's success. |
| 764 | * |
| 765 | * -- wli |
| 766 | */ |
Nick Piggin | 085cc7d | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 767 | static inline void expand(struct zone *zone, struct page *page, |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 768 | int low, int high, struct free_area *area, |
| 769 | int migratetype) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 770 | { |
| 771 | unsigned long size = 1 << high; |
| 772 | |
| 773 | while (high > low) { |
| 774 | area--; |
| 775 | high--; |
| 776 | size >>= 1; |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 777 | VM_BUG_ON(bad_range(zone, &page[size])); |
Stanislaw Gruszka | c0a32fc5 | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 778 | |
| 779 | #ifdef CONFIG_DEBUG_PAGEALLOC |
| 780 | if (high < debug_guardpage_minorder()) { |
| 781 | /* |
| 782 | * Mark as guard pages (or page), that will allow to |
| 783 | * merge back to allocator when buddy will be freed. |
| 784 | * Corresponding page table entries will not be touched, |
| 785 | * pages will stay not present in virtual address space |
| 786 | */ |
| 787 | INIT_LIST_HEAD(&page[size].lru); |
| 788 | set_page_guard_flag(&page[size]); |
| 789 | set_page_private(&page[size], high); |
| 790 | /* Guard pages are not available for any usage */ |
| 791 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << high)); |
| 792 | continue; |
| 793 | } |
| 794 | #endif |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 795 | list_add(&page[size].lru, &area->free_list[migratetype]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 796 | area->nr_free++; |
| 797 | set_page_order(&page[size], high); |
| 798 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 799 | } |
| 800 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 801 | /* |
| 802 | * This page is about to be returned from the page allocator |
| 803 | */ |
Wu Fengguang | 2a7684a | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 804 | static inline int check_new_page(struct page *page) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 805 | { |
Nick Piggin | 92be2e33 | 2006-01-06 00:10:57 -0800 | [diff] [blame] | 806 | if (unlikely(page_mapcount(page) | |
| 807 | (page->mapping != NULL) | |
Nick Piggin | a3af9c3 | 2009-06-16 15:32:10 -0700 | [diff] [blame] | 808 | (atomic_read(&page->_count) != 0) | |
Daisuke Nishimura | f212ad7 | 2011-03-23 16:42:25 -0700 | [diff] [blame] | 809 | (page->flags & PAGE_FLAGS_CHECK_AT_PREP) | |
| 810 | (mem_cgroup_bad_page_check(page)))) { |
Nick Piggin | 224abf9 | 2006-01-06 00:11:11 -0800 | [diff] [blame] | 811 | bad_page(page); |
Hugh Dickins | 689bceb | 2005-11-21 21:32:20 -0800 | [diff] [blame] | 812 | return 1; |
Hugh Dickins | 8cc3b39 | 2009-01-06 14:40:06 -0800 | [diff] [blame] | 813 | } |
Wu Fengguang | 2a7684a | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 814 | return 0; |
| 815 | } |
| 816 | |
| 817 | static int prep_new_page(struct page *page, int order, gfp_t gfp_flags) |
| 818 | { |
| 819 | int i; |
| 820 | |
| 821 | for (i = 0; i < (1 << order); i++) { |
| 822 | struct page *p = page + i; |
| 823 | if (unlikely(check_new_page(p))) |
| 824 | return 1; |
| 825 | } |
Hugh Dickins | 689bceb | 2005-11-21 21:32:20 -0800 | [diff] [blame] | 826 | |
Hugh Dickins | 4c21e2f | 2005-10-29 18:16:40 -0700 | [diff] [blame] | 827 | set_page_private(page, 0); |
Nick Piggin | 7835e98 | 2006-03-22 00:08:40 -0800 | [diff] [blame] | 828 | set_page_refcounted(page); |
Nick Piggin | cc102509 | 2006-12-06 20:32:00 -0800 | [diff] [blame] | 829 | |
| 830 | arch_alloc_page(page, order); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 831 | kernel_map_pages(page, 1 << order, 1); |
Nick Piggin | 17cf440 | 2006-03-22 00:08:41 -0800 | [diff] [blame] | 832 | |
| 833 | if (gfp_flags & __GFP_ZERO) |
| 834 | prep_zero_page(page, order, gfp_flags); |
| 835 | |
| 836 | if (order && (gfp_flags & __GFP_COMP)) |
| 837 | prep_compound_page(page, order); |
| 838 | |
Hugh Dickins | 689bceb | 2005-11-21 21:32:20 -0800 | [diff] [blame] | 839 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 840 | } |
| 841 | |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 842 | /* |
| 843 | * Go through the free lists for the given migratetype and remove |
| 844 | * the smallest available page from the freelists |
| 845 | */ |
Mel Gorman | 728ec98 | 2009-06-16 15:32:04 -0700 | [diff] [blame] | 846 | static inline |
| 847 | struct page *__rmqueue_smallest(struct zone *zone, unsigned int order, |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 848 | int migratetype) |
| 849 | { |
| 850 | unsigned int current_order; |
| 851 | struct free_area * area; |
| 852 | struct page *page; |
| 853 | |
| 854 | /* Find a page of the appropriate size in the preferred list */ |
| 855 | for (current_order = order; current_order < MAX_ORDER; ++current_order) { |
| 856 | area = &(zone->free_area[current_order]); |
| 857 | if (list_empty(&area->free_list[migratetype])) |
| 858 | continue; |
| 859 | |
| 860 | page = list_entry(area->free_list[migratetype].next, |
| 861 | struct page, lru); |
| 862 | list_del(&page->lru); |
| 863 | rmv_page_order(page); |
| 864 | area->nr_free--; |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 865 | expand(zone, page, order, current_order, area, migratetype); |
| 866 | return page; |
| 867 | } |
| 868 | |
| 869 | return NULL; |
| 870 | } |
| 871 | |
| 872 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 873 | /* |
| 874 | * This array describes the order lists are fallen back to when |
| 875 | * the free lists for the desirable migrate type are depleted |
| 876 | */ |
| 877 | static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = { |
Mel Gorman | 64c5e13 | 2007-10-16 01:25:59 -0700 | [diff] [blame] | 878 | [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, |
| 879 | [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_RESERVE }, |
| 880 | [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE }, |
| 881 | [MIGRATE_RESERVE] = { MIGRATE_RESERVE, MIGRATE_RESERVE, MIGRATE_RESERVE }, /* Never used */ |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 882 | }; |
| 883 | |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 884 | /* |
| 885 | * Move the free pages in a range to the free lists of the requested type. |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 886 | * Note that start_page and end_pages are not aligned on a pageblock |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 887 | * boundary. If alignment is required, use move_freepages_block() |
| 888 | */ |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 889 | static int move_freepages(struct zone *zone, |
| 890 | struct page *start_page, struct page *end_page, |
| 891 | int migratetype) |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 892 | { |
| 893 | struct page *page; |
| 894 | unsigned long order; |
Mel Gorman | d100313 | 2007-10-16 01:26:00 -0700 | [diff] [blame] | 895 | int pages_moved = 0; |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 896 | |
| 897 | #ifndef CONFIG_HOLES_IN_ZONE |
| 898 | /* |
| 899 | * page_zone is not safe to call in this context when |
| 900 | * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant |
| 901 | * anyway as we check zone boundaries in move_freepages_block(). |
| 902 | * Remove at a later date when no bug reports exist related to |
Mel Gorman | ac0e5b7 | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 903 | * grouping pages by mobility |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 904 | */ |
| 905 | BUG_ON(page_zone(start_page) != page_zone(end_page)); |
| 906 | #endif |
| 907 | |
| 908 | for (page = start_page; page <= end_page;) { |
Adam Litke | 344c790 | 2008-09-02 14:35:38 -0700 | [diff] [blame] | 909 | /* Make sure we are not inadvertently changing nodes */ |
| 910 | VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone)); |
| 911 | |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 912 | if (!pfn_valid_within(page_to_pfn(page))) { |
| 913 | page++; |
| 914 | continue; |
| 915 | } |
| 916 | |
| 917 | if (!PageBuddy(page)) { |
| 918 | page++; |
| 919 | continue; |
| 920 | } |
| 921 | |
| 922 | order = page_order(page); |
Kirill A. Shutemov | 84be48d | 2011-03-22 16:33:41 -0700 | [diff] [blame] | 923 | list_move(&page->lru, |
| 924 | &zone->free_area[order].free_list[migratetype]); |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 925 | page += 1 << order; |
Mel Gorman | d100313 | 2007-10-16 01:26:00 -0700 | [diff] [blame] | 926 | pages_moved += 1 << order; |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 927 | } |
| 928 | |
Mel Gorman | d100313 | 2007-10-16 01:26:00 -0700 | [diff] [blame] | 929 | return pages_moved; |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 930 | } |
| 931 | |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 932 | static int move_freepages_block(struct zone *zone, struct page *page, |
| 933 | int migratetype) |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 934 | { |
| 935 | unsigned long start_pfn, end_pfn; |
| 936 | struct page *start_page, *end_page; |
| 937 | |
| 938 | start_pfn = page_to_pfn(page); |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 939 | start_pfn = start_pfn & ~(pageblock_nr_pages-1); |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 940 | start_page = pfn_to_page(start_pfn); |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 941 | end_page = start_page + pageblock_nr_pages - 1; |
| 942 | end_pfn = start_pfn + pageblock_nr_pages - 1; |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 943 | |
| 944 | /* Do not cross zone boundaries */ |
| 945 | if (start_pfn < zone->zone_start_pfn) |
| 946 | start_page = page; |
| 947 | if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages) |
| 948 | return 0; |
| 949 | |
| 950 | return move_freepages(zone, start_page, end_page, migratetype); |
| 951 | } |
| 952 | |
Mel Gorman | 2f66a68 | 2009-09-21 17:02:31 -0700 | [diff] [blame] | 953 | static void change_pageblock_range(struct page *pageblock_page, |
| 954 | int start_order, int migratetype) |
| 955 | { |
| 956 | int nr_pageblocks = 1 << (start_order - pageblock_order); |
| 957 | |
| 958 | while (nr_pageblocks--) { |
| 959 | set_pageblock_migratetype(pageblock_page, migratetype); |
| 960 | pageblock_page += pageblock_nr_pages; |
| 961 | } |
| 962 | } |
| 963 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 964 | /* Remove an element from the buddy allocator from the fallback list */ |
Mel Gorman | 0ac3a40 | 2009-06-16 15:32:06 -0700 | [diff] [blame] | 965 | static inline struct page * |
| 966 | __rmqueue_fallback(struct zone *zone, int order, int start_migratetype) |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 967 | { |
| 968 | struct free_area * area; |
| 969 | int current_order; |
| 970 | struct page *page; |
| 971 | int migratetype, i; |
| 972 | |
| 973 | /* Find the largest possible block of pages in the other list */ |
| 974 | for (current_order = MAX_ORDER-1; current_order >= order; |
| 975 | --current_order) { |
| 976 | for (i = 0; i < MIGRATE_TYPES - 1; i++) { |
| 977 | migratetype = fallbacks[start_migratetype][i]; |
| 978 | |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 979 | /* MIGRATE_RESERVE handled later if necessary */ |
| 980 | if (migratetype == MIGRATE_RESERVE) |
| 981 | continue; |
Mel Gorman | e010487 | 2007-10-16 01:25:53 -0700 | [diff] [blame] | 982 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 983 | area = &(zone->free_area[current_order]); |
| 984 | if (list_empty(&area->free_list[migratetype])) |
| 985 | continue; |
| 986 | |
| 987 | page = list_entry(area->free_list[migratetype].next, |
| 988 | struct page, lru); |
| 989 | area->nr_free--; |
| 990 | |
| 991 | /* |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 992 | * If breaking a large block of pages, move all free |
Mel Gorman | 46dafbc | 2007-10-16 01:25:55 -0700 | [diff] [blame] | 993 | * pages to the preferred allocation list. If falling |
| 994 | * back for a reclaimable kernel allocation, be more |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 995 | * aggressive about taking ownership of free pages |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 996 | */ |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 997 | if (unlikely(current_order >= (pageblock_order >> 1)) || |
Mel Gorman | dd5d241 | 2009-09-05 11:17:11 -0700 | [diff] [blame] | 998 | start_migratetype == MIGRATE_RECLAIMABLE || |
| 999 | page_group_by_mobility_disabled) { |
Mel Gorman | 46dafbc | 2007-10-16 01:25:55 -0700 | [diff] [blame] | 1000 | unsigned long pages; |
| 1001 | pages = move_freepages_block(zone, page, |
| 1002 | start_migratetype); |
| 1003 | |
| 1004 | /* Claim the whole block if over half of it is free */ |
Mel Gorman | dd5d241 | 2009-09-05 11:17:11 -0700 | [diff] [blame] | 1005 | if (pages >= (1 << (pageblock_order-1)) || |
| 1006 | page_group_by_mobility_disabled) |
Mel Gorman | 46dafbc | 2007-10-16 01:25:55 -0700 | [diff] [blame] | 1007 | set_pageblock_migratetype(page, |
| 1008 | start_migratetype); |
| 1009 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1010 | migratetype = start_migratetype; |
Mel Gorman | c361be5 | 2007-10-16 01:25:51 -0700 | [diff] [blame] | 1011 | } |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1012 | |
| 1013 | /* Remove the page from the freelists */ |
| 1014 | list_del(&page->lru); |
| 1015 | rmv_page_order(page); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1016 | |
Mel Gorman | 2f66a68 | 2009-09-21 17:02:31 -0700 | [diff] [blame] | 1017 | /* Take ownership for orders >= pageblock_order */ |
| 1018 | if (current_order >= pageblock_order) |
| 1019 | change_pageblock_range(page, current_order, |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1020 | start_migratetype); |
| 1021 | |
| 1022 | expand(zone, page, order, current_order, area, migratetype); |
Mel Gorman | e0fff1b | 2009-09-21 17:02:42 -0700 | [diff] [blame] | 1023 | |
| 1024 | trace_mm_page_alloc_extfrag(page, order, current_order, |
| 1025 | start_migratetype, migratetype); |
| 1026 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1027 | return page; |
| 1028 | } |
| 1029 | } |
| 1030 | |
Mel Gorman | 728ec98 | 2009-06-16 15:32:04 -0700 | [diff] [blame] | 1031 | return NULL; |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1032 | } |
| 1033 | |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 1034 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1035 | * Do the hard work of removing an element from the buddy allocator. |
| 1036 | * Call me with the zone->lock already held. |
| 1037 | */ |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1038 | static struct page *__rmqueue(struct zone *zone, unsigned int order, |
| 1039 | int migratetype) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1040 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1041 | struct page *page; |
| 1042 | |
Mel Gorman | 728ec98 | 2009-06-16 15:32:04 -0700 | [diff] [blame] | 1043 | retry_reserve: |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 1044 | page = __rmqueue_smallest(zone, order, migratetype); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1045 | |
Mel Gorman | 728ec98 | 2009-06-16 15:32:04 -0700 | [diff] [blame] | 1046 | if (unlikely(!page) && migratetype != MIGRATE_RESERVE) { |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 1047 | page = __rmqueue_fallback(zone, order, migratetype); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1048 | |
Mel Gorman | 728ec98 | 2009-06-16 15:32:04 -0700 | [diff] [blame] | 1049 | /* |
| 1050 | * Use MIGRATE_RESERVE rather than fail an allocation. goto |
| 1051 | * is used because __rmqueue_smallest is an inline function |
| 1052 | * and we want just one call site |
| 1053 | */ |
| 1054 | if (!page) { |
| 1055 | migratetype = MIGRATE_RESERVE; |
| 1056 | goto retry_reserve; |
| 1057 | } |
| 1058 | } |
| 1059 | |
Mel Gorman | 0d3d062 | 2009-09-21 17:02:44 -0700 | [diff] [blame] | 1060 | trace_mm_page_alloc_zone_locked(page, order, migratetype); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1061 | return page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1062 | } |
| 1063 | |
| 1064 | /* |
| 1065 | * Obtain a specified number of elements from the buddy allocator, all under |
| 1066 | * a single hold of the lock, for efficiency. Add them to the supplied list. |
| 1067 | * Returns the number of new pages which were placed at *list. |
| 1068 | */ |
| 1069 | static int rmqueue_bulk(struct zone *zone, unsigned int order, |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1070 | unsigned long count, struct list_head *list, |
Mel Gorman | e084b2d | 2009-07-29 15:02:04 -0700 | [diff] [blame] | 1071 | int migratetype, int cold) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1072 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1073 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1074 | |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 1075 | spin_lock(&zone->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1076 | for (i = 0; i < count; ++i) { |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1077 | struct page *page = __rmqueue(zone, order, migratetype); |
Nick Piggin | 085cc7d | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 1078 | if (unlikely(page == NULL)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1079 | break; |
Mel Gorman | 81eabcb | 2007-12-17 16:20:05 -0800 | [diff] [blame] | 1080 | |
| 1081 | /* |
| 1082 | * Split buddy pages returned by expand() are received here |
| 1083 | * in physical page order. The page is added to the callers and |
| 1084 | * list and the list head then moves forward. From the callers |
| 1085 | * perspective, the linked list is ordered by page number in |
| 1086 | * some conditions. This is useful for IO devices that can |
| 1087 | * merge IO requests if the physical pages are ordered |
| 1088 | * properly. |
| 1089 | */ |
Mel Gorman | e084b2d | 2009-07-29 15:02:04 -0700 | [diff] [blame] | 1090 | if (likely(cold == 0)) |
| 1091 | list_add(&page->lru, list); |
| 1092 | else |
| 1093 | list_add_tail(&page->lru, list); |
Mel Gorman | 535131e6 | 2007-10-16 01:25:49 -0700 | [diff] [blame] | 1094 | set_page_private(page, migratetype); |
Mel Gorman | 81eabcb | 2007-12-17 16:20:05 -0800 | [diff] [blame] | 1095 | list = &page->lru; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1096 | } |
Mel Gorman | f2260e6 | 2009-06-16 15:32:13 -0700 | [diff] [blame] | 1097 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 1098 | spin_unlock(&zone->lock); |
Nick Piggin | 085cc7d | 2006-01-06 00:11:01 -0800 | [diff] [blame] | 1099 | return i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1100 | } |
| 1101 | |
Christoph Lameter | 4ae7c03 | 2005-06-21 17:14:57 -0700 | [diff] [blame] | 1102 | #ifdef CONFIG_NUMA |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 1103 | /* |
Christoph Lameter | 4037d45 | 2007-05-09 02:35:14 -0700 | [diff] [blame] | 1104 | * Called from the vmstat counter updater to drain pagesets of this |
| 1105 | * currently executing processor on remote nodes after they have |
| 1106 | * expired. |
| 1107 | * |
Christoph Lameter | 879336c | 2006-03-22 00:09:08 -0800 | [diff] [blame] | 1108 | * Note that this function must be called with the thread pinned to |
| 1109 | * a single processor. |
Christoph Lameter | 8fce4d8 | 2006-03-09 17:33:54 -0800 | [diff] [blame] | 1110 | */ |
Christoph Lameter | 4037d45 | 2007-05-09 02:35:14 -0700 | [diff] [blame] | 1111 | void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp) |
Christoph Lameter | 4ae7c03 | 2005-06-21 17:14:57 -0700 | [diff] [blame] | 1112 | { |
Christoph Lameter | 4ae7c03 | 2005-06-21 17:14:57 -0700 | [diff] [blame] | 1113 | unsigned long flags; |
Christoph Lameter | 4037d45 | 2007-05-09 02:35:14 -0700 | [diff] [blame] | 1114 | int to_drain; |
Christoph Lameter | 4ae7c03 | 2005-06-21 17:14:57 -0700 | [diff] [blame] | 1115 | |
Christoph Lameter | 4037d45 | 2007-05-09 02:35:14 -0700 | [diff] [blame] | 1116 | local_irq_save(flags); |
| 1117 | if (pcp->count >= pcp->batch) |
| 1118 | to_drain = pcp->batch; |
| 1119 | else |
| 1120 | to_drain = pcp->count; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1121 | free_pcppages_bulk(zone, to_drain, pcp); |
Christoph Lameter | 4037d45 | 2007-05-09 02:35:14 -0700 | [diff] [blame] | 1122 | pcp->count -= to_drain; |
| 1123 | local_irq_restore(flags); |
Christoph Lameter | 4ae7c03 | 2005-06-21 17:14:57 -0700 | [diff] [blame] | 1124 | } |
| 1125 | #endif |
| 1126 | |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 1127 | /* |
| 1128 | * Drain pages of the indicated processor. |
| 1129 | * |
| 1130 | * The processor must either be the current processor and the |
| 1131 | * thread pinned to the current processor or a processor that |
| 1132 | * is not online. |
| 1133 | */ |
| 1134 | static void drain_pages(unsigned int cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1135 | { |
Nick Piggin | c54ad30 | 2006-01-06 00:10:56 -0800 | [diff] [blame] | 1136 | unsigned long flags; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1137 | struct zone *zone; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1138 | |
KOSAKI Motohiro | ee99c71 | 2009-03-31 15:19:31 -0700 | [diff] [blame] | 1139 | for_each_populated_zone(zone) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1140 | struct per_cpu_pageset *pset; |
Christoph Lameter | 3dfa572 | 2008-02-04 22:29:19 -0800 | [diff] [blame] | 1141 | struct per_cpu_pages *pcp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1142 | |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 1143 | local_irq_save(flags); |
| 1144 | pset = per_cpu_ptr(zone->pageset, cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1145 | |
Christoph Lameter | 3dfa572 | 2008-02-04 22:29:19 -0800 | [diff] [blame] | 1146 | pcp = &pset->pcp; |
David Rientjes | 2ff754f | 2011-01-25 15:07:23 -0800 | [diff] [blame] | 1147 | if (pcp->count) { |
| 1148 | free_pcppages_bulk(zone, pcp->count, pcp); |
| 1149 | pcp->count = 0; |
| 1150 | } |
Christoph Lameter | 3dfa572 | 2008-02-04 22:29:19 -0800 | [diff] [blame] | 1151 | local_irq_restore(flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1152 | } |
| 1153 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1154 | |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 1155 | /* |
| 1156 | * Spill all of this CPU's per-cpu pages back into the buddy allocator. |
| 1157 | */ |
| 1158 | void drain_local_pages(void *arg) |
| 1159 | { |
| 1160 | drain_pages(smp_processor_id()); |
| 1161 | } |
| 1162 | |
| 1163 | /* |
| 1164 | * Spill all the per-cpu pages from all CPUs back into the buddy allocator |
| 1165 | */ |
| 1166 | void drain_all_pages(void) |
| 1167 | { |
Jens Axboe | 15c8b6c | 2008-05-09 09:39:44 +0200 | [diff] [blame] | 1168 | on_each_cpu(drain_local_pages, NULL, 1); |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 1169 | } |
| 1170 | |
Rafael J. Wysocki | 296699d | 2007-07-29 23:27:18 +0200 | [diff] [blame] | 1171 | #ifdef CONFIG_HIBERNATION |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1172 | |
| 1173 | void mark_free_pages(struct zone *zone) |
| 1174 | { |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 1175 | unsigned long pfn, max_zone_pfn; |
| 1176 | unsigned long flags; |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1177 | int order, t; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1178 | struct list_head *curr; |
| 1179 | |
| 1180 | if (!zone->spanned_pages) |
| 1181 | return; |
| 1182 | |
| 1183 | spin_lock_irqsave(&zone->lock, flags); |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 1184 | |
| 1185 | max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages; |
| 1186 | for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++) |
| 1187 | if (pfn_valid(pfn)) { |
| 1188 | struct page *page = pfn_to_page(pfn); |
| 1189 | |
Rafael J. Wysocki | 7be9823 | 2007-05-06 14:50:42 -0700 | [diff] [blame] | 1190 | if (!swsusp_page_is_forbidden(page)) |
| 1191 | swsusp_unset_page_free(page); |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 1192 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1193 | |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1194 | for_each_migratetype_order(order, t) { |
| 1195 | list_for_each(curr, &zone->free_area[order].free_list[t]) { |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 1196 | unsigned long i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1197 | |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 1198 | pfn = page_to_pfn(list_entry(curr, struct page, lru)); |
| 1199 | for (i = 0; i < (1UL << order); i++) |
Rafael J. Wysocki | 7be9823 | 2007-05-06 14:50:42 -0700 | [diff] [blame] | 1200 | swsusp_set_page_free(pfn_to_page(pfn + i)); |
Rafael J. Wysocki | f623f0d | 2006-09-25 23:32:49 -0700 | [diff] [blame] | 1201 | } |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1202 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1203 | spin_unlock_irqrestore(&zone->lock, flags); |
| 1204 | } |
Mel Gorman | e2c55dc | 2007-10-16 01:25:50 -0700 | [diff] [blame] | 1205 | #endif /* CONFIG_PM */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1206 | |
| 1207 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1208 | * Free a 0-order page |
Li Hong | fc91668 | 2010-03-05 13:41:54 -0800 | [diff] [blame] | 1209 | * cold == 1 ? free a cold page : free a hot page |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1210 | */ |
Li Hong | fc91668 | 2010-03-05 13:41:54 -0800 | [diff] [blame] | 1211 | void free_hot_cold_page(struct page *page, int cold) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1212 | { |
| 1213 | struct zone *zone = page_zone(page); |
| 1214 | struct per_cpu_pages *pcp; |
| 1215 | unsigned long flags; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1216 | int migratetype; |
Johannes Weiner | 451ea25 | 2009-09-21 17:01:48 -0700 | [diff] [blame] | 1217 | int wasMlocked = __TestClearPageMlocked(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1218 | |
KOSAKI Motohiro | ec95f53 | 2010-05-24 14:32:38 -0700 | [diff] [blame] | 1219 | if (!free_pages_prepare(page, 0)) |
Hugh Dickins | 689bceb | 2005-11-21 21:32:20 -0800 | [diff] [blame] | 1220 | return; |
| 1221 | |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1222 | migratetype = get_pageblock_migratetype(page); |
| 1223 | set_page_private(page, migratetype); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1224 | local_irq_save(flags); |
Johannes Weiner | c277331 | 2009-06-19 19:30:56 +0200 | [diff] [blame] | 1225 | if (unlikely(wasMlocked)) |
Mel Gorman | da456f1 | 2009-06-16 15:32:08 -0700 | [diff] [blame] | 1226 | free_page_mlock(page); |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 1227 | __count_vm_event(PGFREE); |
Mel Gorman | da456f1 | 2009-06-16 15:32:08 -0700 | [diff] [blame] | 1228 | |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1229 | /* |
| 1230 | * We only track unmovable, reclaimable and movable on pcp lists. |
| 1231 | * Free ISOLATE pages back to the allocator because they are being |
| 1232 | * offlined but treat RESERVE as movable pages so we can get those |
| 1233 | * areas back if necessary. Otherwise, we may have to free |
| 1234 | * excessively into the page allocator |
| 1235 | */ |
| 1236 | if (migratetype >= MIGRATE_PCPTYPES) { |
| 1237 | if (unlikely(migratetype == MIGRATE_ISOLATE)) { |
| 1238 | free_one_page(zone, page, 0, migratetype); |
| 1239 | goto out; |
| 1240 | } |
| 1241 | migratetype = MIGRATE_MOVABLE; |
| 1242 | } |
| 1243 | |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 1244 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
Christoph Lameter | 3dfa572 | 2008-02-04 22:29:19 -0800 | [diff] [blame] | 1245 | if (cold) |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1246 | list_add_tail(&page->lru, &pcp->lists[migratetype]); |
Christoph Lameter | 3dfa572 | 2008-02-04 22:29:19 -0800 | [diff] [blame] | 1247 | else |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1248 | list_add(&page->lru, &pcp->lists[migratetype]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1249 | pcp->count++; |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 1250 | if (pcp->count >= pcp->high) { |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1251 | free_pcppages_bulk(zone, pcp->batch, pcp); |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 1252 | pcp->count -= pcp->batch; |
| 1253 | } |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1254 | |
| 1255 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1256 | local_irq_restore(flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1257 | } |
| 1258 | |
Nick Piggin | 8dfcc9b | 2006-03-22 00:08:05 -0800 | [diff] [blame] | 1259 | /* |
Konstantin Khlebnikov | cc59850 | 2012-01-10 15:07:04 -0800 | [diff] [blame] | 1260 | * Free a list of 0-order pages |
| 1261 | */ |
| 1262 | void free_hot_cold_page_list(struct list_head *list, int cold) |
| 1263 | { |
| 1264 | struct page *page, *next; |
| 1265 | |
| 1266 | list_for_each_entry_safe(page, next, list, lru) { |
Konstantin Khlebnikov | b413d48 | 2012-01-10 15:07:09 -0800 | [diff] [blame] | 1267 | trace_mm_page_free_batched(page, cold); |
Konstantin Khlebnikov | cc59850 | 2012-01-10 15:07:04 -0800 | [diff] [blame] | 1268 | free_hot_cold_page(page, cold); |
| 1269 | } |
| 1270 | } |
| 1271 | |
| 1272 | /* |
Nick Piggin | 8dfcc9b | 2006-03-22 00:08:05 -0800 | [diff] [blame] | 1273 | * split_page takes a non-compound higher-order page, and splits it into |
| 1274 | * n (1<<order) sub-pages: page[0..n] |
| 1275 | * Each sub-page must be freed individually. |
| 1276 | * |
| 1277 | * Note: this is probably too low level an operation for use in drivers. |
| 1278 | * Please consult with lkml before using this in your driver. |
| 1279 | */ |
| 1280 | void split_page(struct page *page, unsigned int order) |
| 1281 | { |
| 1282 | int i; |
| 1283 | |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 1284 | VM_BUG_ON(PageCompound(page)); |
| 1285 | VM_BUG_ON(!page_count(page)); |
Vegard Nossum | b1eeab6 | 2008-11-25 16:55:53 +0100 | [diff] [blame] | 1286 | |
| 1287 | #ifdef CONFIG_KMEMCHECK |
| 1288 | /* |
| 1289 | * Split shadow pages too, because free(page[0]) would |
| 1290 | * otherwise free the whole shadow. |
| 1291 | */ |
| 1292 | if (kmemcheck_page_is_tracked(page)) |
| 1293 | split_page(virt_to_page(page[0].shadow), order); |
| 1294 | #endif |
| 1295 | |
Nick Piggin | 7835e98 | 2006-03-22 00:08:40 -0800 | [diff] [blame] | 1296 | for (i = 1; i < (1 << order); i++) |
| 1297 | set_page_refcounted(page + i); |
Nick Piggin | 8dfcc9b | 2006-03-22 00:08:05 -0800 | [diff] [blame] | 1298 | } |
Nick Piggin | 8dfcc9b | 2006-03-22 00:08:05 -0800 | [diff] [blame] | 1299 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1300 | /* |
Mel Gorman | 748446b | 2010-05-24 14:32:27 -0700 | [diff] [blame] | 1301 | * Similar to split_page except the page is already free. As this is only |
| 1302 | * being used for migration, the migratetype of the block also changes. |
| 1303 | * As this is called with interrupts disabled, the caller is responsible |
| 1304 | * for calling arch_alloc_page() and kernel_map_page() after interrupts |
| 1305 | * are enabled. |
| 1306 | * |
| 1307 | * Note: this is probably too low level an operation for use in drivers. |
| 1308 | * Please consult with lkml before using this in your driver. |
| 1309 | */ |
| 1310 | int split_free_page(struct page *page) |
| 1311 | { |
| 1312 | unsigned int order; |
| 1313 | unsigned long watermark; |
| 1314 | struct zone *zone; |
| 1315 | |
| 1316 | BUG_ON(!PageBuddy(page)); |
| 1317 | |
| 1318 | zone = page_zone(page); |
| 1319 | order = page_order(page); |
| 1320 | |
| 1321 | /* Obey watermarks as if the page was being allocated */ |
| 1322 | watermark = low_wmark_pages(zone) + (1 << order); |
| 1323 | if (!zone_watermark_ok(zone, 0, watermark, 0, 0)) |
| 1324 | return 0; |
| 1325 | |
| 1326 | /* Remove page from free list */ |
| 1327 | list_del(&page->lru); |
| 1328 | zone->free_area[order].nr_free--; |
| 1329 | rmv_page_order(page); |
| 1330 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(1UL << order)); |
| 1331 | |
| 1332 | /* Split into individual pages */ |
| 1333 | set_page_refcounted(page); |
| 1334 | split_page(page, order); |
| 1335 | |
| 1336 | if (order >= pageblock_order - 1) { |
| 1337 | struct page *endpage = page + (1 << order) - 1; |
| 1338 | for (; page < endpage; page += pageblock_nr_pages) |
| 1339 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
| 1340 | } |
| 1341 | |
| 1342 | return 1 << order; |
| 1343 | } |
| 1344 | |
| 1345 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1346 | * Really, prep_compound_page() should be called from __rmqueue_bulk(). But |
| 1347 | * we cheat by calling it from here, in the order > 0 path. Saves a branch |
| 1348 | * or two. |
| 1349 | */ |
Mel Gorman | 0a15c3e | 2009-06-16 15:32:05 -0700 | [diff] [blame] | 1350 | static inline |
| 1351 | struct page *buffered_rmqueue(struct zone *preferred_zone, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 1352 | struct zone *zone, int order, gfp_t gfp_flags, |
| 1353 | int migratetype) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1354 | { |
| 1355 | unsigned long flags; |
Hugh Dickins | 689bceb | 2005-11-21 21:32:20 -0800 | [diff] [blame] | 1356 | struct page *page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1357 | int cold = !!(gfp_flags & __GFP_COLD); |
| 1358 | |
Hugh Dickins | 689bceb | 2005-11-21 21:32:20 -0800 | [diff] [blame] | 1359 | again: |
Nick Piggin | 48db57f | 2006-01-08 01:00:42 -0800 | [diff] [blame] | 1360 | if (likely(order == 0)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1361 | struct per_cpu_pages *pcp; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1362 | struct list_head *list; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1363 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1364 | local_irq_save(flags); |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 1365 | pcp = &this_cpu_ptr(zone->pageset)->pcp; |
| 1366 | list = &pcp->lists[migratetype]; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1367 | if (list_empty(list)) { |
Mel Gorman | 535131e6 | 2007-10-16 01:25:49 -0700 | [diff] [blame] | 1368 | pcp->count += rmqueue_bulk(zone, 0, |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1369 | pcp->batch, list, |
Mel Gorman | e084b2d | 2009-07-29 15:02:04 -0700 | [diff] [blame] | 1370 | migratetype, cold); |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1371 | if (unlikely(list_empty(list))) |
Shaohua Li | 6fb332f | 2009-09-21 17:01:17 -0700 | [diff] [blame] | 1372 | goto failed; |
Mel Gorman | 535131e6 | 2007-10-16 01:25:49 -0700 | [diff] [blame] | 1373 | } |
Mel Gorman | b92a6ed | 2007-10-16 01:25:50 -0700 | [diff] [blame] | 1374 | |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 1375 | if (cold) |
| 1376 | page = list_entry(list->prev, struct page, lru); |
| 1377 | else |
| 1378 | page = list_entry(list->next, struct page, lru); |
| 1379 | |
Mel Gorman | b92a6ed | 2007-10-16 01:25:50 -0700 | [diff] [blame] | 1380 | list_del(&page->lru); |
| 1381 | pcp->count--; |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1382 | } else { |
Andrew Morton | dab48da | 2009-06-16 15:32:37 -0700 | [diff] [blame] | 1383 | if (unlikely(gfp_flags & __GFP_NOFAIL)) { |
| 1384 | /* |
| 1385 | * __GFP_NOFAIL is not to be used in new code. |
| 1386 | * |
| 1387 | * All __GFP_NOFAIL callers should be fixed so that they |
| 1388 | * properly detect and handle allocation failures. |
| 1389 | * |
| 1390 | * We most definitely don't want callers attempting to |
Linus Torvalds | 4923abf | 2009-06-24 12:16:49 -0700 | [diff] [blame] | 1391 | * allocate greater than order-1 page units with |
Andrew Morton | dab48da | 2009-06-16 15:32:37 -0700 | [diff] [blame] | 1392 | * __GFP_NOFAIL. |
| 1393 | */ |
Linus Torvalds | 4923abf | 2009-06-24 12:16:49 -0700 | [diff] [blame] | 1394 | WARN_ON_ONCE(order > 1); |
Andrew Morton | dab48da | 2009-06-16 15:32:37 -0700 | [diff] [blame] | 1395 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1396 | spin_lock_irqsave(&zone->lock, flags); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 1397 | page = __rmqueue(zone, order, migratetype); |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 1398 | spin_unlock(&zone->lock); |
| 1399 | if (!page) |
| 1400 | goto failed; |
KOSAKI Motohiro | 6ccf80e | 2010-01-15 17:01:18 -0800 | [diff] [blame] | 1401 | __mod_zone_page_state(zone, NR_FREE_PAGES, -(1 << order)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1402 | } |
| 1403 | |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 1404 | __count_zone_vm_events(PGALLOC, zone, 1 << order); |
Andi Kleen | 78afd56 | 2011-03-22 16:33:12 -0700 | [diff] [blame] | 1405 | zone_statistics(preferred_zone, zone, gfp_flags); |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 1406 | local_irq_restore(flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1407 | |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 1408 | VM_BUG_ON(bad_range(zone, page)); |
Nick Piggin | 17cf440 | 2006-03-22 00:08:41 -0800 | [diff] [blame] | 1409 | if (prep_new_page(page, order, gfp_flags)) |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 1410 | goto again; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1411 | return page; |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 1412 | |
| 1413 | failed: |
| 1414 | local_irq_restore(flags); |
Nick Piggin | a74609f | 2006-01-06 00:11:20 -0800 | [diff] [blame] | 1415 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1416 | } |
| 1417 | |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 1418 | /* The ALLOC_WMARK bits are used as an index to zone->watermark */ |
| 1419 | #define ALLOC_WMARK_MIN WMARK_MIN |
| 1420 | #define ALLOC_WMARK_LOW WMARK_LOW |
| 1421 | #define ALLOC_WMARK_HIGH WMARK_HIGH |
| 1422 | #define ALLOC_NO_WATERMARKS 0x04 /* don't check watermarks at all */ |
| 1423 | |
| 1424 | /* Mask to get the watermark bits */ |
| 1425 | #define ALLOC_WMARK_MASK (ALLOC_NO_WATERMARKS-1) |
| 1426 | |
Nick Piggin | 3148890 | 2005-11-28 13:44:03 -0800 | [diff] [blame] | 1427 | #define ALLOC_HARDER 0x10 /* try to alloc harder */ |
| 1428 | #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */ |
| 1429 | #define ALLOC_CPUSET 0x40 /* check for correct cpuset */ |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1430 | |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 1431 | #ifdef CONFIG_FAIL_PAGE_ALLOC |
| 1432 | |
Akinobu Mita | b2588c4 | 2011-07-26 16:09:03 -0700 | [diff] [blame] | 1433 | static struct { |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 1434 | struct fault_attr attr; |
| 1435 | |
| 1436 | u32 ignore_gfp_highmem; |
| 1437 | u32 ignore_gfp_wait; |
Akinobu Mita | 5411499 | 2007-07-15 23:40:23 -0700 | [diff] [blame] | 1438 | u32 min_order; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 1439 | } fail_page_alloc = { |
| 1440 | .attr = FAULT_ATTR_INITIALIZER, |
Don Mullis | 6b1b60f | 2006-12-08 02:39:53 -0800 | [diff] [blame] | 1441 | .ignore_gfp_wait = 1, |
| 1442 | .ignore_gfp_highmem = 1, |
Akinobu Mita | 5411499 | 2007-07-15 23:40:23 -0700 | [diff] [blame] | 1443 | .min_order = 1, |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 1444 | }; |
| 1445 | |
| 1446 | static int __init setup_fail_page_alloc(char *str) |
| 1447 | { |
| 1448 | return setup_fault_attr(&fail_page_alloc.attr, str); |
| 1449 | } |
| 1450 | __setup("fail_page_alloc=", setup_fail_page_alloc); |
| 1451 | |
| 1452 | static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
| 1453 | { |
Akinobu Mita | 5411499 | 2007-07-15 23:40:23 -0700 | [diff] [blame] | 1454 | if (order < fail_page_alloc.min_order) |
| 1455 | return 0; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 1456 | if (gfp_mask & __GFP_NOFAIL) |
| 1457 | return 0; |
| 1458 | if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM)) |
| 1459 | return 0; |
| 1460 | if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT)) |
| 1461 | return 0; |
| 1462 | |
| 1463 | return should_fail(&fail_page_alloc.attr, 1 << order); |
| 1464 | } |
| 1465 | |
| 1466 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS |
| 1467 | |
| 1468 | static int __init fail_page_alloc_debugfs(void) |
| 1469 | { |
Al Viro | f4ae40a | 2011-07-24 04:33:43 -0400 | [diff] [blame] | 1470 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 1471 | struct dentry *dir; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 1472 | |
Akinobu Mita | dd48c08 | 2011-08-03 16:21:01 -0700 | [diff] [blame] | 1473 | dir = fault_create_debugfs_attr("fail_page_alloc", NULL, |
| 1474 | &fail_page_alloc.attr); |
| 1475 | if (IS_ERR(dir)) |
| 1476 | return PTR_ERR(dir); |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 1477 | |
Akinobu Mita | b2588c4 | 2011-07-26 16:09:03 -0700 | [diff] [blame] | 1478 | if (!debugfs_create_bool("ignore-gfp-wait", mode, dir, |
| 1479 | &fail_page_alloc.ignore_gfp_wait)) |
| 1480 | goto fail; |
| 1481 | if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir, |
| 1482 | &fail_page_alloc.ignore_gfp_highmem)) |
| 1483 | goto fail; |
| 1484 | if (!debugfs_create_u32("min-order", mode, dir, |
| 1485 | &fail_page_alloc.min_order)) |
| 1486 | goto fail; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 1487 | |
Akinobu Mita | b2588c4 | 2011-07-26 16:09:03 -0700 | [diff] [blame] | 1488 | return 0; |
| 1489 | fail: |
Akinobu Mita | dd48c08 | 2011-08-03 16:21:01 -0700 | [diff] [blame] | 1490 | debugfs_remove_recursive(dir); |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 1491 | |
Akinobu Mita | b2588c4 | 2011-07-26 16:09:03 -0700 | [diff] [blame] | 1492 | return -ENOMEM; |
Akinobu Mita | 933e312 | 2006-12-08 02:39:45 -0800 | [diff] [blame] | 1493 | } |
| 1494 | |
| 1495 | late_initcall(fail_page_alloc_debugfs); |
| 1496 | |
| 1497 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ |
| 1498 | |
| 1499 | #else /* CONFIG_FAIL_PAGE_ALLOC */ |
| 1500 | |
| 1501 | static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order) |
| 1502 | { |
| 1503 | return 0; |
| 1504 | } |
| 1505 | |
| 1506 | #endif /* CONFIG_FAIL_PAGE_ALLOC */ |
| 1507 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1508 | /* |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 1509 | * Return true if free pages are above 'mark'. This takes into account the order |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1510 | * of the allocation. |
| 1511 | */ |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 1512 | static bool __zone_watermark_ok(struct zone *z, int order, unsigned long mark, |
| 1513 | int classzone_idx, int alloc_flags, long free_pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1514 | { |
| 1515 | /* free_pages my go negative - that's OK */ |
Christoph Lameter | d23ad42 | 2007-02-10 01:43:02 -0800 | [diff] [blame] | 1516 | long min = mark; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1517 | int o; |
| 1518 | |
Michal Hocko | df0a6da | 2012-01-10 15:08:02 -0800 | [diff] [blame] | 1519 | free_pages -= (1 << order) - 1; |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1520 | if (alloc_flags & ALLOC_HIGH) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1521 | min -= min / 2; |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1522 | if (alloc_flags & ALLOC_HARDER) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1523 | min -= min / 4; |
| 1524 | |
| 1525 | if (free_pages <= min + z->lowmem_reserve[classzone_idx]) |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 1526 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1527 | for (o = 0; o < order; o++) { |
| 1528 | /* At the next order, this order's pages become unavailable */ |
| 1529 | free_pages -= z->free_area[o].nr_free << o; |
| 1530 | |
| 1531 | /* Require fewer higher order pages to be free */ |
| 1532 | min >>= 1; |
| 1533 | |
| 1534 | if (free_pages <= min) |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 1535 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1536 | } |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 1537 | return true; |
| 1538 | } |
| 1539 | |
| 1540 | bool zone_watermark_ok(struct zone *z, int order, unsigned long mark, |
| 1541 | int classzone_idx, int alloc_flags) |
| 1542 | { |
| 1543 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, |
| 1544 | zone_page_state(z, NR_FREE_PAGES)); |
| 1545 | } |
| 1546 | |
| 1547 | bool zone_watermark_ok_safe(struct zone *z, int order, unsigned long mark, |
| 1548 | int classzone_idx, int alloc_flags) |
| 1549 | { |
| 1550 | long free_pages = zone_page_state(z, NR_FREE_PAGES); |
| 1551 | |
| 1552 | if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) |
| 1553 | free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES); |
| 1554 | |
| 1555 | return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags, |
| 1556 | free_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1557 | } |
| 1558 | |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1559 | #ifdef CONFIG_NUMA |
| 1560 | /* |
| 1561 | * zlc_setup - Setup for "zonelist cache". Uses cached zone data to |
| 1562 | * skip over zones that are not allowed by the cpuset, or that have |
| 1563 | * been recently (in last second) found to be nearly full. See further |
| 1564 | * comments in mmzone.h. Reduces cache footprint of zonelist scans |
Simon Arlott | 183ff22 | 2007-10-20 01:27:18 +0200 | [diff] [blame] | 1565 | * that have to skip over a lot of full or unallowed zones. |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1566 | * |
| 1567 | * If the zonelist cache is present in the passed in zonelist, then |
| 1568 | * returns a pointer to the allowed node mask (either the current |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 1569 | * tasks mems_allowed, or node_states[N_HIGH_MEMORY].) |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1570 | * |
| 1571 | * If the zonelist cache is not available for this zonelist, does |
| 1572 | * nothing and returns NULL. |
| 1573 | * |
| 1574 | * If the fullzones BITMAP in the zonelist cache is stale (more than |
| 1575 | * a second since last zap'd) then we zap it out (clear its bits.) |
| 1576 | * |
| 1577 | * We hold off even calling zlc_setup, until after we've checked the |
| 1578 | * first zone in the zonelist, on the theory that most allocations will |
| 1579 | * be satisfied from that first zone, so best to examine that zone as |
| 1580 | * quickly as we can. |
| 1581 | */ |
| 1582 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) |
| 1583 | { |
| 1584 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ |
| 1585 | nodemask_t *allowednodes; /* zonelist_cache approximation */ |
| 1586 | |
| 1587 | zlc = zonelist->zlcache_ptr; |
| 1588 | if (!zlc) |
| 1589 | return NULL; |
| 1590 | |
S.Caglar Onur | f05111f | 2008-04-28 02:12:38 -0700 | [diff] [blame] | 1591 | if (time_after(jiffies, zlc->last_full_zap + HZ)) { |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1592 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); |
| 1593 | zlc->last_full_zap = jiffies; |
| 1594 | } |
| 1595 | |
| 1596 | allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ? |
| 1597 | &cpuset_current_mems_allowed : |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 1598 | &node_states[N_HIGH_MEMORY]; |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1599 | return allowednodes; |
| 1600 | } |
| 1601 | |
| 1602 | /* |
| 1603 | * Given 'z' scanning a zonelist, run a couple of quick checks to see |
| 1604 | * if it is worth looking at further for free memory: |
| 1605 | * 1) Check that the zone isn't thought to be full (doesn't have its |
| 1606 | * bit set in the zonelist_cache fullzones BITMAP). |
| 1607 | * 2) Check that the zones node (obtained from the zonelist_cache |
| 1608 | * z_to_n[] mapping) is allowed in the passed in allowednodes mask. |
| 1609 | * Return true (non-zero) if zone is worth looking at further, or |
| 1610 | * else return false (zero) if it is not. |
| 1611 | * |
| 1612 | * This check -ignores- the distinction between various watermarks, |
| 1613 | * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is |
| 1614 | * found to be full for any variation of these watermarks, it will |
| 1615 | * be considered full for up to one second by all requests, unless |
| 1616 | * we are so low on memory on all allowed nodes that we are forced |
| 1617 | * into the second scan of the zonelist. |
| 1618 | * |
| 1619 | * In the second scan we ignore this zonelist cache and exactly |
| 1620 | * apply the watermarks to all zones, even it is slower to do so. |
| 1621 | * We are low on memory in the second scan, and should leave no stone |
| 1622 | * unturned looking for a free page. |
| 1623 | */ |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1624 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1625 | nodemask_t *allowednodes) |
| 1626 | { |
| 1627 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ |
| 1628 | int i; /* index of *z in zonelist zones */ |
| 1629 | int n; /* node that zone *z is on */ |
| 1630 | |
| 1631 | zlc = zonelist->zlcache_ptr; |
| 1632 | if (!zlc) |
| 1633 | return 1; |
| 1634 | |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1635 | i = z - zonelist->_zonerefs; |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1636 | n = zlc->z_to_n[i]; |
| 1637 | |
| 1638 | /* This zone is worth trying if it is allowed but not full */ |
| 1639 | return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones); |
| 1640 | } |
| 1641 | |
| 1642 | /* |
| 1643 | * Given 'z' scanning a zonelist, set the corresponding bit in |
| 1644 | * zlc->fullzones, so that subsequent attempts to allocate a page |
| 1645 | * from that zone don't waste time re-examining it. |
| 1646 | */ |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1647 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1648 | { |
| 1649 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ |
| 1650 | int i; /* index of *z in zonelist zones */ |
| 1651 | |
| 1652 | zlc = zonelist->zlcache_ptr; |
| 1653 | if (!zlc) |
| 1654 | return; |
| 1655 | |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1656 | i = z - zonelist->_zonerefs; |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1657 | |
| 1658 | set_bit(i, zlc->fullzones); |
| 1659 | } |
| 1660 | |
Mel Gorman | 76d3fbf | 2011-07-25 17:12:30 -0700 | [diff] [blame] | 1661 | /* |
| 1662 | * clear all zones full, called after direct reclaim makes progress so that |
| 1663 | * a zone that was recently full is not skipped over for up to a second |
| 1664 | */ |
| 1665 | static void zlc_clear_zones_full(struct zonelist *zonelist) |
| 1666 | { |
| 1667 | struct zonelist_cache *zlc; /* cached zonelist speedup info */ |
| 1668 | |
| 1669 | zlc = zonelist->zlcache_ptr; |
| 1670 | if (!zlc) |
| 1671 | return; |
| 1672 | |
| 1673 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); |
| 1674 | } |
| 1675 | |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1676 | #else /* CONFIG_NUMA */ |
| 1677 | |
| 1678 | static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags) |
| 1679 | { |
| 1680 | return NULL; |
| 1681 | } |
| 1682 | |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1683 | static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z, |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1684 | nodemask_t *allowednodes) |
| 1685 | { |
| 1686 | return 1; |
| 1687 | } |
| 1688 | |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1689 | static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z) |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1690 | { |
| 1691 | } |
Mel Gorman | 76d3fbf | 2011-07-25 17:12:30 -0700 | [diff] [blame] | 1692 | |
| 1693 | static void zlc_clear_zones_full(struct zonelist *zonelist) |
| 1694 | { |
| 1695 | } |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1696 | #endif /* CONFIG_NUMA */ |
| 1697 | |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1698 | /* |
Paul Jackson | 0798e51 | 2006-12-06 20:31:38 -0800 | [diff] [blame] | 1699 | * get_page_from_freelist goes through the zonelist trying to allocate |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1700 | * a page. |
| 1701 | */ |
| 1702 | static struct page * |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 1703 | get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order, |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 1704 | struct zonelist *zonelist, int high_zoneidx, int alloc_flags, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 1705 | struct zone *preferred_zone, int migratetype) |
Martin Hicks | 753ee72 | 2005-06-21 17:14:41 -0700 | [diff] [blame] | 1706 | { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 1707 | struct zoneref *z; |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1708 | struct page *page = NULL; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 1709 | int classzone_idx; |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 1710 | struct zone *zone; |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1711 | nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */ |
| 1712 | int zlc_active = 0; /* set if using zonelist_cache */ |
| 1713 | int did_zlc_setup = 0; /* just call zlc_setup() one time */ |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 1714 | |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 1715 | classzone_idx = zone_idx(preferred_zone); |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1716 | zonelist_scan: |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1717 | /* |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1718 | * Scan zonelist, looking for a zone with enough free. |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1719 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
| 1720 | */ |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 1721 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
| 1722 | high_zoneidx, nodemask) { |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1723 | if (NUMA_BUILD && zlc_active && |
| 1724 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) |
| 1725 | continue; |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1726 | if ((alloc_flags & ALLOC_CPUSET) && |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 1727 | !cpuset_zone_allowed_softwall(zone, gfp_mask)) |
Mel Gorman | cd38b11 | 2011-07-25 17:12:29 -0700 | [diff] [blame] | 1728 | continue; |
Johannes Weiner | a756cf5 | 2012-01-10 15:07:49 -0800 | [diff] [blame] | 1729 | /* |
| 1730 | * When allocating a page cache page for writing, we |
| 1731 | * want to get it from a zone that is within its dirty |
| 1732 | * limit, such that no single zone holds more than its |
| 1733 | * proportional share of globally allowed dirty pages. |
| 1734 | * The dirty limits take into account the zone's |
| 1735 | * lowmem reserves and high watermark so that kswapd |
| 1736 | * should be able to balance it without having to |
| 1737 | * write pages from its LRU list. |
| 1738 | * |
| 1739 | * This may look like it could increase pressure on |
| 1740 | * lower zones by failing allocations in higher zones |
| 1741 | * before they are full. But the pages that do spill |
| 1742 | * over are limited as the lower zones are protected |
| 1743 | * by this very same mechanism. It should not become |
| 1744 | * a practical burden to them. |
| 1745 | * |
| 1746 | * XXX: For now, allow allocations to potentially |
| 1747 | * exceed the per-zone dirty limit in the slowpath |
| 1748 | * (ALLOC_WMARK_LOW unset) before going into reclaim, |
| 1749 | * which is important when on a NUMA setup the allowed |
| 1750 | * zones are together not big enough to reach the |
| 1751 | * global limit. The proper fix for these situations |
| 1752 | * will require awareness of zones in the |
| 1753 | * dirty-throttling and the flusher threads. |
| 1754 | */ |
| 1755 | if ((alloc_flags & ALLOC_WMARK_LOW) && |
| 1756 | (gfp_mask & __GFP_WRITE) && !zone_dirty_ok(zone)) |
| 1757 | goto this_zone_full; |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1758 | |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 1759 | BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK); |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1760 | if (!(alloc_flags & ALLOC_NO_WATERMARKS)) { |
Nick Piggin | 3148890 | 2005-11-28 13:44:03 -0800 | [diff] [blame] | 1761 | unsigned long mark; |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 1762 | int ret; |
| 1763 | |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 1764 | mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK]; |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 1765 | if (zone_watermark_ok(zone, order, mark, |
| 1766 | classzone_idx, alloc_flags)) |
| 1767 | goto try_this_zone; |
| 1768 | |
Mel Gorman | cd38b11 | 2011-07-25 17:12:29 -0700 | [diff] [blame] | 1769 | if (NUMA_BUILD && !did_zlc_setup && nr_online_nodes > 1) { |
| 1770 | /* |
| 1771 | * we do zlc_setup if there are multiple nodes |
| 1772 | * and before considering the first zone allowed |
| 1773 | * by the cpuset. |
| 1774 | */ |
| 1775 | allowednodes = zlc_setup(zonelist, alloc_flags); |
| 1776 | zlc_active = 1; |
| 1777 | did_zlc_setup = 1; |
| 1778 | } |
| 1779 | |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 1780 | if (zone_reclaim_mode == 0) |
| 1781 | goto this_zone_full; |
| 1782 | |
Mel Gorman | cd38b11 | 2011-07-25 17:12:29 -0700 | [diff] [blame] | 1783 | /* |
| 1784 | * As we may have just activated ZLC, check if the first |
| 1785 | * eligible zone has failed zone_reclaim recently. |
| 1786 | */ |
| 1787 | if (NUMA_BUILD && zlc_active && |
| 1788 | !zlc_zone_worth_trying(zonelist, z, allowednodes)) |
| 1789 | continue; |
| 1790 | |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 1791 | ret = zone_reclaim(zone, gfp_mask, order); |
| 1792 | switch (ret) { |
| 1793 | case ZONE_RECLAIM_NOSCAN: |
| 1794 | /* did not scan */ |
Mel Gorman | cd38b11 | 2011-07-25 17:12:29 -0700 | [diff] [blame] | 1795 | continue; |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 1796 | case ZONE_RECLAIM_FULL: |
| 1797 | /* scanned but unreclaimable */ |
Mel Gorman | cd38b11 | 2011-07-25 17:12:29 -0700 | [diff] [blame] | 1798 | continue; |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 1799 | default: |
| 1800 | /* did we reclaim enough */ |
| 1801 | if (!zone_watermark_ok(zone, order, mark, |
| 1802 | classzone_idx, alloc_flags)) |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1803 | goto this_zone_full; |
Paul Jackson | 0798e51 | 2006-12-06 20:31:38 -0800 | [diff] [blame] | 1804 | } |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1805 | } |
| 1806 | |
Mel Gorman | fa5e084 | 2009-06-16 15:33:22 -0700 | [diff] [blame] | 1807 | try_this_zone: |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 1808 | page = buffered_rmqueue(preferred_zone, zone, order, |
| 1809 | gfp_mask, migratetype); |
Paul Jackson | 0798e51 | 2006-12-06 20:31:38 -0800 | [diff] [blame] | 1810 | if (page) |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1811 | break; |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1812 | this_zone_full: |
| 1813 | if (NUMA_BUILD) |
| 1814 | zlc_mark_zone_full(zonelist, z); |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 1815 | } |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 1816 | |
| 1817 | if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) { |
| 1818 | /* Disable zlc cache for second zonelist scan */ |
| 1819 | zlc_active = 0; |
| 1820 | goto zonelist_scan; |
| 1821 | } |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1822 | return page; |
Martin Hicks | 753ee72 | 2005-06-21 17:14:41 -0700 | [diff] [blame] | 1823 | } |
| 1824 | |
David Rientjes | 29423e77 | 2011-03-22 16:30:47 -0700 | [diff] [blame] | 1825 | /* |
| 1826 | * Large machines with many possible nodes should not always dump per-node |
| 1827 | * meminfo in irq context. |
| 1828 | */ |
| 1829 | static inline bool should_suppress_show_mem(void) |
| 1830 | { |
| 1831 | bool ret = false; |
| 1832 | |
| 1833 | #if NODES_SHIFT > 8 |
| 1834 | ret = in_interrupt(); |
| 1835 | #endif |
| 1836 | return ret; |
| 1837 | } |
| 1838 | |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 1839 | static DEFINE_RATELIMIT_STATE(nopage_rs, |
| 1840 | DEFAULT_RATELIMIT_INTERVAL, |
| 1841 | DEFAULT_RATELIMIT_BURST); |
| 1842 | |
| 1843 | void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...) |
| 1844 | { |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 1845 | unsigned int filter = SHOW_MEM_FILTER_NODES; |
| 1846 | |
Stanislaw Gruszka | c0a32fc5 | 2012-01-10 15:07:28 -0800 | [diff] [blame] | 1847 | if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) || |
| 1848 | debug_guardpage_minorder() > 0) |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 1849 | return; |
| 1850 | |
| 1851 | /* |
| 1852 | * This documents exceptions given to allocations in certain |
| 1853 | * contexts that are allowed to allocate outside current's set |
| 1854 | * of allowed nodes. |
| 1855 | */ |
| 1856 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
| 1857 | if (test_thread_flag(TIF_MEMDIE) || |
| 1858 | (current->flags & (PF_MEMALLOC | PF_EXITING))) |
| 1859 | filter &= ~SHOW_MEM_FILTER_NODES; |
| 1860 | if (in_interrupt() || !(gfp_mask & __GFP_WAIT)) |
| 1861 | filter &= ~SHOW_MEM_FILTER_NODES; |
| 1862 | |
| 1863 | if (fmt) { |
Joe Perches | 3ee9a4f | 2011-10-31 17:08:35 -0700 | [diff] [blame] | 1864 | struct va_format vaf; |
| 1865 | va_list args; |
| 1866 | |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 1867 | va_start(args, fmt); |
Joe Perches | 3ee9a4f | 2011-10-31 17:08:35 -0700 | [diff] [blame] | 1868 | |
| 1869 | vaf.fmt = fmt; |
| 1870 | vaf.va = &args; |
| 1871 | |
| 1872 | pr_warn("%pV", &vaf); |
| 1873 | |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 1874 | va_end(args); |
| 1875 | } |
| 1876 | |
Joe Perches | 3ee9a4f | 2011-10-31 17:08:35 -0700 | [diff] [blame] | 1877 | pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n", |
| 1878 | current->comm, order, gfp_mask); |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 1879 | |
| 1880 | dump_stack(); |
| 1881 | if (!should_suppress_show_mem()) |
| 1882 | show_mem(filter); |
| 1883 | } |
| 1884 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1885 | static inline int |
| 1886 | should_alloc_retry(gfp_t gfp_mask, unsigned int order, |
Mel Gorman | f90ac39 | 2012-01-10 15:07:15 -0800 | [diff] [blame] | 1887 | unsigned long did_some_progress, |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1888 | unsigned long pages_reclaimed) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1889 | { |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1890 | /* Do not loop if specifically requested */ |
| 1891 | if (gfp_mask & __GFP_NORETRY) |
| 1892 | return 0; |
| 1893 | |
Mel Gorman | f90ac39 | 2012-01-10 15:07:15 -0800 | [diff] [blame] | 1894 | /* Always retry if specifically requested */ |
| 1895 | if (gfp_mask & __GFP_NOFAIL) |
| 1896 | return 1; |
| 1897 | |
| 1898 | /* |
| 1899 | * Suspend converts GFP_KERNEL to __GFP_WAIT which can prevent reclaim |
| 1900 | * making forward progress without invoking OOM. Suspend also disables |
| 1901 | * storage devices so kswapd will not help. Bail if we are suspending. |
| 1902 | */ |
| 1903 | if (!did_some_progress && pm_suspended_storage()) |
| 1904 | return 0; |
| 1905 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1906 | /* |
| 1907 | * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER |
| 1908 | * means __GFP_NOFAIL, but that may not be true in other |
| 1909 | * implementations. |
| 1910 | */ |
| 1911 | if (order <= PAGE_ALLOC_COSTLY_ORDER) |
| 1912 | return 1; |
| 1913 | |
| 1914 | /* |
| 1915 | * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is |
| 1916 | * specified, then we retry until we no longer reclaim any pages |
| 1917 | * (above), or we've reclaimed an order of pages at least as |
| 1918 | * large as the allocation's order. In both cases, if the |
| 1919 | * allocation still fails, we stop retrying. |
| 1920 | */ |
| 1921 | if (gfp_mask & __GFP_REPEAT && pages_reclaimed < (1 << order)) |
| 1922 | return 1; |
| 1923 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1924 | return 0; |
| 1925 | } |
| 1926 | |
| 1927 | static inline struct page * |
| 1928 | __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order, |
| 1929 | struct zonelist *zonelist, enum zone_type high_zoneidx, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 1930 | nodemask_t *nodemask, struct zone *preferred_zone, |
| 1931 | int migratetype) |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1932 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1933 | struct page *page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1934 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1935 | /* Acquire the OOM killer lock for the zones in zonelist */ |
Minchan Kim | ff321fe | 2010-08-09 17:18:57 -0700 | [diff] [blame] | 1936 | if (!try_set_zonelist_oom(zonelist, gfp_mask)) { |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1937 | schedule_timeout_uninterruptible(1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1938 | return NULL; |
| 1939 | } |
Jens Axboe | 6b1de91 | 2005-11-17 21:35:02 +0100 | [diff] [blame] | 1940 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1941 | /* |
| 1942 | * Go through the zonelist yet one more time, keep very high watermark |
| 1943 | * here, this is only to catch a parallel oom killing, we must fail if |
| 1944 | * we're still under heavy pressure. |
| 1945 | */ |
| 1946 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, |
| 1947 | order, zonelist, high_zoneidx, |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 1948 | ALLOC_WMARK_HIGH|ALLOC_CPUSET, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 1949 | preferred_zone, migratetype); |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 1950 | if (page) |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1951 | goto out; |
| 1952 | |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 1953 | if (!(gfp_mask & __GFP_NOFAIL)) { |
| 1954 | /* The OOM killer will not help higher order allocs */ |
| 1955 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
| 1956 | goto out; |
David Rientjes | 03668b3 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 1957 | /* The OOM killer does not needlessly kill tasks for lowmem */ |
| 1958 | if (high_zoneidx < ZONE_NORMAL) |
| 1959 | goto out; |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 1960 | /* |
| 1961 | * GFP_THISNODE contains __GFP_NORETRY and we never hit this. |
| 1962 | * Sanity check for bare calls of __GFP_THISNODE, not real OOM. |
| 1963 | * The caller should handle page allocation failure by itself if |
| 1964 | * it specifies __GFP_THISNODE. |
| 1965 | * Note: Hugepage uses it but will hit PAGE_ALLOC_COSTLY_ORDER. |
| 1966 | */ |
| 1967 | if (gfp_mask & __GFP_THISNODE) |
| 1968 | goto out; |
| 1969 | } |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1970 | /* Exhausted what can be done so it's blamo time */ |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 1971 | out_of_memory(zonelist, gfp_mask, order, nodemask); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 1972 | |
| 1973 | out: |
| 1974 | clear_zonelist_oom(zonelist, gfp_mask); |
| 1975 | return page; |
| 1976 | } |
| 1977 | |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1978 | #ifdef CONFIG_COMPACTION |
| 1979 | /* Try memory compaction for high-order allocations before reclaim */ |
| 1980 | static struct page * |
| 1981 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, |
| 1982 | struct zonelist *zonelist, enum zone_type high_zoneidx, |
| 1983 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, |
Mel Gorman | 6619971 | 2012-01-12 17:19:41 -0800 | [diff] [blame] | 1984 | int migratetype, bool sync_migration, |
| 1985 | bool *deferred_compaction, |
| 1986 | unsigned long *did_some_progress) |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1987 | { |
| 1988 | struct page *page; |
| 1989 | |
Mel Gorman | 6619971 | 2012-01-12 17:19:41 -0800 | [diff] [blame] | 1990 | if (!order) |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1991 | return NULL; |
| 1992 | |
Mel Gorman | 6619971 | 2012-01-12 17:19:41 -0800 | [diff] [blame] | 1993 | if (compaction_deferred(preferred_zone)) { |
| 1994 | *deferred_compaction = true; |
| 1995 | return NULL; |
| 1996 | } |
| 1997 | |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 1998 | current->flags |= PF_MEMALLOC; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 1999 | *did_some_progress = try_to_compact_pages(zonelist, order, gfp_mask, |
Mel Gorman | 77f1fe6 | 2011-01-13 15:45:57 -0800 | [diff] [blame] | 2000 | nodemask, sync_migration); |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 2001 | current->flags &= ~PF_MEMALLOC; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 2002 | if (*did_some_progress != COMPACT_SKIPPED) { |
| 2003 | |
| 2004 | /* Page migration frees to the PCP lists but we want merging */ |
| 2005 | drain_pages(get_cpu()); |
| 2006 | put_cpu(); |
| 2007 | |
| 2008 | page = get_page_from_freelist(gfp_mask, nodemask, |
| 2009 | order, zonelist, high_zoneidx, |
| 2010 | alloc_flags, preferred_zone, |
| 2011 | migratetype); |
| 2012 | if (page) { |
Mel Gorman | 4f92e25 | 2010-05-24 14:32:32 -0700 | [diff] [blame] | 2013 | preferred_zone->compact_considered = 0; |
| 2014 | preferred_zone->compact_defer_shift = 0; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 2015 | count_vm_event(COMPACTSUCCESS); |
| 2016 | return page; |
| 2017 | } |
| 2018 | |
| 2019 | /* |
| 2020 | * It's bad if compaction run occurs and fails. |
| 2021 | * The most likely reason is that pages exist, |
| 2022 | * but not enough to satisfy watermarks. |
| 2023 | */ |
| 2024 | count_vm_event(COMPACTFAIL); |
Mel Gorman | 6619971 | 2012-01-12 17:19:41 -0800 | [diff] [blame] | 2025 | |
| 2026 | /* |
| 2027 | * As async compaction considers a subset of pageblocks, only |
| 2028 | * defer if the failure was a sync compaction failure. |
| 2029 | */ |
| 2030 | if (sync_migration) |
| 2031 | defer_compaction(preferred_zone); |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 2032 | |
| 2033 | cond_resched(); |
| 2034 | } |
| 2035 | |
| 2036 | return NULL; |
| 2037 | } |
| 2038 | #else |
| 2039 | static inline struct page * |
| 2040 | __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order, |
| 2041 | struct zonelist *zonelist, enum zone_type high_zoneidx, |
| 2042 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, |
Mel Gorman | 6619971 | 2012-01-12 17:19:41 -0800 | [diff] [blame] | 2043 | int migratetype, bool sync_migration, |
| 2044 | bool *deferred_compaction, |
| 2045 | unsigned long *did_some_progress) |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 2046 | { |
| 2047 | return NULL; |
| 2048 | } |
| 2049 | #endif /* CONFIG_COMPACTION */ |
| 2050 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2051 | /* The really slow allocator path where we enter direct reclaim */ |
| 2052 | static inline struct page * |
| 2053 | __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order, |
| 2054 | struct zonelist *zonelist, enum zone_type high_zoneidx, |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 2055 | nodemask_t *nodemask, int alloc_flags, struct zone *preferred_zone, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 2056 | int migratetype, unsigned long *did_some_progress) |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2057 | { |
| 2058 | struct page *page = NULL; |
| 2059 | struct reclaim_state reclaim_state; |
Mel Gorman | 9ee493c | 2010-09-09 16:38:18 -0700 | [diff] [blame] | 2060 | bool drained = false; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2061 | |
| 2062 | cond_resched(); |
| 2063 | |
| 2064 | /* We now go into synchronous reclaim */ |
| 2065 | cpuset_memory_pressure_bump(); |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 2066 | current->flags |= PF_MEMALLOC; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2067 | lockdep_set_current_reclaim_state(gfp_mask); |
| 2068 | reclaim_state.reclaimed_slab = 0; |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 2069 | current->reclaim_state = &reclaim_state; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2070 | |
| 2071 | *did_some_progress = try_to_free_pages(zonelist, order, gfp_mask, nodemask); |
| 2072 | |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 2073 | current->reclaim_state = NULL; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2074 | lockdep_clear_current_reclaim_state(); |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 2075 | current->flags &= ~PF_MEMALLOC; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2076 | |
| 2077 | cond_resched(); |
| 2078 | |
Mel Gorman | 9ee493c | 2010-09-09 16:38:18 -0700 | [diff] [blame] | 2079 | if (unlikely(!(*did_some_progress))) |
| 2080 | return NULL; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2081 | |
Mel Gorman | 76d3fbf | 2011-07-25 17:12:30 -0700 | [diff] [blame] | 2082 | /* After successful reclaim, reconsider all zones for allocation */ |
| 2083 | if (NUMA_BUILD) |
| 2084 | zlc_clear_zones_full(zonelist); |
| 2085 | |
Mel Gorman | 9ee493c | 2010-09-09 16:38:18 -0700 | [diff] [blame] | 2086 | retry: |
| 2087 | page = get_page_from_freelist(gfp_mask, nodemask, order, |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 2088 | zonelist, high_zoneidx, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 2089 | alloc_flags, preferred_zone, |
| 2090 | migratetype); |
Mel Gorman | 9ee493c | 2010-09-09 16:38:18 -0700 | [diff] [blame] | 2091 | |
| 2092 | /* |
| 2093 | * If an allocation failed after direct reclaim, it could be because |
| 2094 | * pages are pinned on the per-cpu lists. Drain them and try again |
| 2095 | */ |
| 2096 | if (!page && !drained) { |
| 2097 | drain_all_pages(); |
| 2098 | drained = true; |
| 2099 | goto retry; |
| 2100 | } |
| 2101 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2102 | return page; |
| 2103 | } |
| 2104 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2105 | /* |
| 2106 | * This is called in the allocator slow-path if the allocation request is of |
| 2107 | * sufficient urgency to ignore watermarks and take other desperate measures |
| 2108 | */ |
| 2109 | static inline struct page * |
| 2110 | __alloc_pages_high_priority(gfp_t gfp_mask, unsigned int order, |
| 2111 | struct zonelist *zonelist, enum zone_type high_zoneidx, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 2112 | nodemask_t *nodemask, struct zone *preferred_zone, |
| 2113 | int migratetype) |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2114 | { |
| 2115 | struct page *page; |
| 2116 | |
| 2117 | do { |
| 2118 | page = get_page_from_freelist(gfp_mask, nodemask, order, |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 2119 | zonelist, high_zoneidx, ALLOC_NO_WATERMARKS, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 2120 | preferred_zone, migratetype); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2121 | |
| 2122 | if (!page && gfp_mask & __GFP_NOFAIL) |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 2123 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2124 | } while (!page && (gfp_mask & __GFP_NOFAIL)); |
| 2125 | |
| 2126 | return page; |
| 2127 | } |
| 2128 | |
| 2129 | static inline |
| 2130 | void wake_all_kswapd(unsigned int order, struct zonelist *zonelist, |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2131 | enum zone_type high_zoneidx, |
| 2132 | enum zone_type classzone_idx) |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2133 | { |
| 2134 | struct zoneref *z; |
| 2135 | struct zone *zone; |
| 2136 | |
| 2137 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2138 | wakeup_kswapd(zone, order, classzone_idx); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2139 | } |
| 2140 | |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 2141 | static inline int |
| 2142 | gfp_to_alloc_flags(gfp_t gfp_mask) |
| 2143 | { |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 2144 | int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET; |
| 2145 | const gfp_t wait = gfp_mask & __GFP_WAIT; |
| 2146 | |
Mel Gorman | a56f57f | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 2147 | /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */ |
Namhyung Kim | e6223a3 | 2010-10-26 14:21:59 -0700 | [diff] [blame] | 2148 | BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH); |
Mel Gorman | a56f57f | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 2149 | |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 2150 | /* |
| 2151 | * The caller may dip into page reserves a bit more if the caller |
| 2152 | * cannot run direct reclaim, or if the caller has realtime scheduling |
| 2153 | * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will |
| 2154 | * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH). |
| 2155 | */ |
Namhyung Kim | e6223a3 | 2010-10-26 14:21:59 -0700 | [diff] [blame] | 2156 | alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH); |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 2157 | |
| 2158 | if (!wait) { |
Andrea Arcangeli | 5c3240d | 2011-01-13 15:46:49 -0800 | [diff] [blame] | 2159 | /* |
| 2160 | * Not worth trying to allocate harder for |
| 2161 | * __GFP_NOMEMALLOC even if it can't schedule. |
| 2162 | */ |
| 2163 | if (!(gfp_mask & __GFP_NOMEMALLOC)) |
| 2164 | alloc_flags |= ALLOC_HARDER; |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 2165 | /* |
| 2166 | * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc. |
| 2167 | * See also cpuset_zone_allowed() comment in kernel/cpuset.c. |
| 2168 | */ |
| 2169 | alloc_flags &= ~ALLOC_CPUSET; |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 2170 | } else if (unlikely(rt_task(current)) && !in_interrupt()) |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 2171 | alloc_flags |= ALLOC_HARDER; |
| 2172 | |
| 2173 | if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) { |
| 2174 | if (!in_interrupt() && |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 2175 | ((current->flags & PF_MEMALLOC) || |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 2176 | unlikely(test_thread_flag(TIF_MEMDIE)))) |
| 2177 | alloc_flags |= ALLOC_NO_WATERMARKS; |
| 2178 | } |
| 2179 | |
| 2180 | return alloc_flags; |
| 2181 | } |
| 2182 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2183 | static inline struct page * |
| 2184 | __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order, |
| 2185 | struct zonelist *zonelist, enum zone_type high_zoneidx, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 2186 | nodemask_t *nodemask, struct zone *preferred_zone, |
| 2187 | int migratetype) |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2188 | { |
| 2189 | const gfp_t wait = gfp_mask & __GFP_WAIT; |
| 2190 | struct page *page = NULL; |
| 2191 | int alloc_flags; |
| 2192 | unsigned long pages_reclaimed = 0; |
| 2193 | unsigned long did_some_progress; |
Mel Gorman | 77f1fe6 | 2011-01-13 15:45:57 -0800 | [diff] [blame] | 2194 | bool sync_migration = false; |
Mel Gorman | 6619971 | 2012-01-12 17:19:41 -0800 | [diff] [blame] | 2195 | bool deferred_compaction = false; |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2196 | |
Christoph Lameter | 952f3b5 | 2006-12-06 20:33:26 -0800 | [diff] [blame] | 2197 | /* |
Mel Gorman | 72807a7 | 2009-06-16 15:32:18 -0700 | [diff] [blame] | 2198 | * In the slowpath, we sanity check order to avoid ever trying to |
| 2199 | * reclaim >= MAX_ORDER areas which will never succeed. Callers may |
| 2200 | * be using allocators in order of preference for an area that is |
| 2201 | * too large. |
| 2202 | */ |
Mel Gorman | 1fc28b7 | 2009-07-29 15:04:08 -0700 | [diff] [blame] | 2203 | if (order >= MAX_ORDER) { |
| 2204 | WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN)); |
Mel Gorman | 72807a7 | 2009-06-16 15:32:18 -0700 | [diff] [blame] | 2205 | return NULL; |
Mel Gorman | 1fc28b7 | 2009-07-29 15:04:08 -0700 | [diff] [blame] | 2206 | } |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2207 | |
Christoph Lameter | 952f3b5 | 2006-12-06 20:33:26 -0800 | [diff] [blame] | 2208 | /* |
| 2209 | * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and |
| 2210 | * __GFP_NOWARN set) should not cause reclaim since the subsystem |
| 2211 | * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim |
| 2212 | * using a larger set of nodes after it has established that the |
| 2213 | * allowed per node queues are empty and that nodes are |
| 2214 | * over allocated. |
| 2215 | */ |
| 2216 | if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE) |
| 2217 | goto nopage; |
| 2218 | |
Mel Gorman | cc4a685 | 2009-11-11 14:26:14 -0800 | [diff] [blame] | 2219 | restart: |
Andrea Arcangeli | 32dba98 | 2011-01-13 15:46:49 -0800 | [diff] [blame] | 2220 | if (!(gfp_mask & __GFP_NO_KSWAPD)) |
| 2221 | wake_all_kswapd(order, zonelist, high_zoneidx, |
Mel Gorman | 9950474 | 2011-01-13 15:46:20 -0800 | [diff] [blame] | 2222 | zone_idx(preferred_zone)); |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2223 | |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2224 | /* |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2225 | * OK, we're below the kswapd watermark and have kicked background |
| 2226 | * reclaim. Now things get more complex, so set up alloc_flags according |
| 2227 | * to how we want to proceed. |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2228 | */ |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 2229 | alloc_flags = gfp_to_alloc_flags(gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2230 | |
David Rientjes | f33261d | 2011-01-25 15:07:20 -0800 | [diff] [blame] | 2231 | /* |
| 2232 | * Find the true preferred zone if the allocation is unconstrained by |
| 2233 | * cpusets. |
| 2234 | */ |
| 2235 | if (!(alloc_flags & ALLOC_CPUSET) && !nodemask) |
| 2236 | first_zones_zonelist(zonelist, high_zoneidx, NULL, |
| 2237 | &preferred_zone); |
| 2238 | |
Andrew Barry | cfa54a0 | 2011-05-24 17:12:52 -0700 | [diff] [blame] | 2239 | rebalance: |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 2240 | /* This is the last chance, in general, before the goto nopage. */ |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 2241 | page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist, |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 2242 | high_zoneidx, alloc_flags & ~ALLOC_NO_WATERMARKS, |
| 2243 | preferred_zone, migratetype); |
Rohit Seth | 7fb1d9f | 2005-11-13 16:06:43 -0800 | [diff] [blame] | 2244 | if (page) |
| 2245 | goto got_pg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2246 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2247 | /* Allocate without watermarks if the context allows */ |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 2248 | if (alloc_flags & ALLOC_NO_WATERMARKS) { |
| 2249 | page = __alloc_pages_high_priority(gfp_mask, order, |
| 2250 | zonelist, high_zoneidx, nodemask, |
| 2251 | preferred_zone, migratetype); |
| 2252 | if (page) |
| 2253 | goto got_pg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2254 | } |
| 2255 | |
| 2256 | /* Atomic allocations - we can't balance anything */ |
| 2257 | if (!wait) |
| 2258 | goto nopage; |
| 2259 | |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 2260 | /* Avoid recursion of direct reclaim */ |
Andrew Morton | c06b1fc | 2011-01-13 15:47:32 -0800 | [diff] [blame] | 2261 | if (current->flags & PF_MEMALLOC) |
Peter Zijlstra | 341ce06 | 2009-06-16 15:32:02 -0700 | [diff] [blame] | 2262 | goto nopage; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2263 | |
David Rientjes | 6583bb6 | 2009-07-29 15:02:06 -0700 | [diff] [blame] | 2264 | /* Avoid allocations with no watermarks from looping endlessly */ |
| 2265 | if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL)) |
| 2266 | goto nopage; |
| 2267 | |
Mel Gorman | 77f1fe6 | 2011-01-13 15:45:57 -0800 | [diff] [blame] | 2268 | /* |
| 2269 | * Try direct compaction. The first pass is asynchronous. Subsequent |
| 2270 | * attempts after direct reclaim are synchronous |
| 2271 | */ |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 2272 | page = __alloc_pages_direct_compact(gfp_mask, order, |
| 2273 | zonelist, high_zoneidx, |
| 2274 | nodemask, |
| 2275 | alloc_flags, preferred_zone, |
Mel Gorman | 6619971 | 2012-01-12 17:19:41 -0800 | [diff] [blame] | 2276 | migratetype, sync_migration, |
| 2277 | &deferred_compaction, |
| 2278 | &did_some_progress); |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 2279 | if (page) |
| 2280 | goto got_pg; |
Andrea Arcangeli | c6a140b | 2011-05-24 17:11:38 -0700 | [diff] [blame] | 2281 | sync_migration = true; |
Mel Gorman | 56de726 | 2010-05-24 14:32:30 -0700 | [diff] [blame] | 2282 | |
Mel Gorman | 6619971 | 2012-01-12 17:19:41 -0800 | [diff] [blame] | 2283 | /* |
| 2284 | * If compaction is deferred for high-order allocations, it is because |
| 2285 | * sync compaction recently failed. In this is the case and the caller |
| 2286 | * has requested the system not be heavily disrupted, fail the |
| 2287 | * allocation now instead of entering direct reclaim |
| 2288 | */ |
| 2289 | if (deferred_compaction && (gfp_mask & __GFP_NO_KSWAPD)) |
| 2290 | goto nopage; |
| 2291 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2292 | /* Try direct reclaim and then allocating */ |
| 2293 | page = __alloc_pages_direct_reclaim(gfp_mask, order, |
| 2294 | zonelist, high_zoneidx, |
| 2295 | nodemask, |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 2296 | alloc_flags, preferred_zone, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 2297 | migratetype, &did_some_progress); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2298 | if (page) |
| 2299 | goto got_pg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2300 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2301 | /* |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2302 | * If we failed to make any progress reclaiming, then we are |
| 2303 | * running out of options and have to consider going OOM |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2304 | */ |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2305 | if (!did_some_progress) { |
| 2306 | if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) { |
Rafael J. Wysocki | 7f33d49 | 2009-06-16 15:32:41 -0700 | [diff] [blame] | 2307 | if (oom_killer_disabled) |
| 2308 | goto nopage; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2309 | page = __alloc_pages_may_oom(gfp_mask, order, |
| 2310 | zonelist, high_zoneidx, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 2311 | nodemask, preferred_zone, |
| 2312 | migratetype); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2313 | if (page) |
| 2314 | goto got_pg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2315 | |
David Rientjes | 03668b3 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 2316 | if (!(gfp_mask & __GFP_NOFAIL)) { |
| 2317 | /* |
| 2318 | * The oom killer is not called for high-order |
| 2319 | * allocations that may fail, so if no progress |
| 2320 | * is being made, there are no other options and |
| 2321 | * retrying is unlikely to help. |
| 2322 | */ |
| 2323 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
| 2324 | goto nopage; |
| 2325 | /* |
| 2326 | * The oom killer is not called for lowmem |
| 2327 | * allocations to prevent needlessly killing |
| 2328 | * innocent tasks. |
| 2329 | */ |
| 2330 | if (high_zoneidx < ZONE_NORMAL) |
| 2331 | goto nopage; |
| 2332 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2333 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2334 | goto restart; |
| 2335 | } |
| 2336 | } |
| 2337 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2338 | /* Check if we should retry the allocation */ |
Nishanth Aravamudan | a41f24e | 2008-04-29 00:58:25 -0700 | [diff] [blame] | 2339 | pages_reclaimed += did_some_progress; |
Mel Gorman | f90ac39 | 2012-01-10 15:07:15 -0800 | [diff] [blame] | 2340 | if (should_alloc_retry(gfp_mask, order, did_some_progress, |
| 2341 | pages_reclaimed)) { |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2342 | /* Wait for some write requests to complete then retry */ |
Mel Gorman | 0e093d99 | 2010-10-26 14:21:45 -0700 | [diff] [blame] | 2343 | wait_iff_congested(preferred_zone, BLK_RW_ASYNC, HZ/50); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2344 | goto rebalance; |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 2345 | } else { |
| 2346 | /* |
| 2347 | * High-order allocations do not necessarily loop after |
| 2348 | * direct reclaim and reclaim/compaction depends on compaction |
| 2349 | * being called after reclaim so call directly if necessary |
| 2350 | */ |
| 2351 | page = __alloc_pages_direct_compact(gfp_mask, order, |
| 2352 | zonelist, high_zoneidx, |
| 2353 | nodemask, |
| 2354 | alloc_flags, preferred_zone, |
Mel Gorman | 6619971 | 2012-01-12 17:19:41 -0800 | [diff] [blame] | 2355 | migratetype, sync_migration, |
| 2356 | &deferred_compaction, |
| 2357 | &did_some_progress); |
Mel Gorman | 3e7d344 | 2011-01-13 15:45:56 -0800 | [diff] [blame] | 2358 | if (page) |
| 2359 | goto got_pg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2360 | } |
| 2361 | |
| 2362 | nopage: |
Dave Hansen | a238ab5 | 2011-05-24 17:12:16 -0700 | [diff] [blame] | 2363 | warn_alloc_failed(gfp_mask, order, NULL); |
Vegard Nossum | b1eeab6 | 2008-11-25 16:55:53 +0100 | [diff] [blame] | 2364 | return page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2365 | got_pg: |
Vegard Nossum | b1eeab6 | 2008-11-25 16:55:53 +0100 | [diff] [blame] | 2366 | if (kmemcheck_enabled) |
| 2367 | kmemcheck_pagealloc_alloc(page, order, gfp_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2368 | return page; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2369 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2370 | } |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2371 | |
| 2372 | /* |
| 2373 | * This is the 'heart' of the zoned buddy allocator. |
| 2374 | */ |
| 2375 | struct page * |
| 2376 | __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, |
| 2377 | struct zonelist *zonelist, nodemask_t *nodemask) |
| 2378 | { |
| 2379 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 2380 | struct zone *preferred_zone; |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2381 | struct page *page; |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 2382 | int migratetype = allocflags_to_migratetype(gfp_mask); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2383 | |
Benjamin Herrenschmidt | dcce284 | 2009-06-18 13:24:12 +1000 | [diff] [blame] | 2384 | gfp_mask &= gfp_allowed_mask; |
| 2385 | |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2386 | lockdep_trace_alloc(gfp_mask); |
| 2387 | |
| 2388 | might_sleep_if(gfp_mask & __GFP_WAIT); |
| 2389 | |
| 2390 | if (should_fail_alloc_page(gfp_mask, order)) |
| 2391 | return NULL; |
| 2392 | |
| 2393 | /* |
| 2394 | * Check the zones suitable for the gfp_mask contain at least one |
| 2395 | * valid zone. It's possible to have an empty zonelist as a result |
| 2396 | * of GFP_THISNODE and a memoryless node |
| 2397 | */ |
| 2398 | if (unlikely(!zonelist->_zonerefs->zone)) |
| 2399 | return NULL; |
| 2400 | |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 2401 | get_mems_allowed(); |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 2402 | /* The preferred zone is used for statistics later */ |
David Rientjes | f33261d | 2011-01-25 15:07:20 -0800 | [diff] [blame] | 2403 | first_zones_zonelist(zonelist, high_zoneidx, |
| 2404 | nodemask ? : &cpuset_current_mems_allowed, |
| 2405 | &preferred_zone); |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 2406 | if (!preferred_zone) { |
| 2407 | put_mems_allowed(); |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 2408 | return NULL; |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 2409 | } |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 2410 | |
| 2411 | /* First allocation attempt */ |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2412 | page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order, |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 2413 | zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 2414 | preferred_zone, migratetype); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2415 | if (unlikely(!page)) |
| 2416 | page = __alloc_pages_slowpath(gfp_mask, order, |
Mel Gorman | 5117f45 | 2009-06-16 15:31:59 -0700 | [diff] [blame] | 2417 | zonelist, high_zoneidx, nodemask, |
Mel Gorman | 3dd2826 | 2009-06-16 15:32:00 -0700 | [diff] [blame] | 2418 | preferred_zone, migratetype); |
Miao Xie | c0ff745 | 2010-05-24 14:32:08 -0700 | [diff] [blame] | 2419 | put_mems_allowed(); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2420 | |
Mel Gorman | 4b4f278 | 2009-09-21 17:02:41 -0700 | [diff] [blame] | 2421 | trace_mm_page_alloc(page, order, gfp_mask, migratetype); |
Mel Gorman | 11e33f6 | 2009-06-16 15:31:57 -0700 | [diff] [blame] | 2422 | return page; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2423 | } |
Mel Gorman | d239171 | 2009-06-16 15:31:52 -0700 | [diff] [blame] | 2424 | EXPORT_SYMBOL(__alloc_pages_nodemask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2425 | |
| 2426 | /* |
| 2427 | * Common helper functions. |
| 2428 | */ |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 2429 | unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2430 | { |
Akinobu Mita | 945a111 | 2009-09-21 17:01:47 -0700 | [diff] [blame] | 2431 | struct page *page; |
| 2432 | |
| 2433 | /* |
| 2434 | * __get_free_pages() returns a 32-bit address, which cannot represent |
| 2435 | * a highmem page |
| 2436 | */ |
| 2437 | VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0); |
| 2438 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2439 | page = alloc_pages(gfp_mask, order); |
| 2440 | if (!page) |
| 2441 | return 0; |
| 2442 | return (unsigned long) page_address(page); |
| 2443 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2444 | EXPORT_SYMBOL(__get_free_pages); |
| 2445 | |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 2446 | unsigned long get_zeroed_page(gfp_t gfp_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2447 | { |
Akinobu Mita | 945a111 | 2009-09-21 17:01:47 -0700 | [diff] [blame] | 2448 | return __get_free_pages(gfp_mask | __GFP_ZERO, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2449 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2450 | EXPORT_SYMBOL(get_zeroed_page); |
| 2451 | |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 2452 | void __free_pages(struct page *page, unsigned int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2453 | { |
Nick Piggin | b581003 | 2005-10-29 18:16:12 -0700 | [diff] [blame] | 2454 | if (put_page_testzero(page)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2455 | if (order == 0) |
Li Hong | fc91668 | 2010-03-05 13:41:54 -0800 | [diff] [blame] | 2456 | free_hot_cold_page(page, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2457 | else |
| 2458 | __free_pages_ok(page, order); |
| 2459 | } |
| 2460 | } |
| 2461 | |
| 2462 | EXPORT_SYMBOL(__free_pages); |
| 2463 | |
Harvey Harrison | 920c7a5 | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 2464 | void free_pages(unsigned long addr, unsigned int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2465 | { |
| 2466 | if (addr != 0) { |
Nick Piggin | 725d704 | 2006-09-25 23:30:55 -0700 | [diff] [blame] | 2467 | VM_BUG_ON(!virt_addr_valid((void *)addr)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2468 | __free_pages(virt_to_page((void *)addr), order); |
| 2469 | } |
| 2470 | } |
| 2471 | |
| 2472 | EXPORT_SYMBOL(free_pages); |
| 2473 | |
Andi Kleen | ee85c2e | 2011-05-11 15:13:34 -0700 | [diff] [blame] | 2474 | static void *make_alloc_exact(unsigned long addr, unsigned order, size_t size) |
| 2475 | { |
| 2476 | if (addr) { |
| 2477 | unsigned long alloc_end = addr + (PAGE_SIZE << order); |
| 2478 | unsigned long used = addr + PAGE_ALIGN(size); |
| 2479 | |
| 2480 | split_page(virt_to_page((void *)addr), order); |
| 2481 | while (used < alloc_end) { |
| 2482 | free_page(used); |
| 2483 | used += PAGE_SIZE; |
| 2484 | } |
| 2485 | } |
| 2486 | return (void *)addr; |
| 2487 | } |
| 2488 | |
Timur Tabi | 2be0ffe | 2008-07-23 21:28:11 -0700 | [diff] [blame] | 2489 | /** |
| 2490 | * alloc_pages_exact - allocate an exact number physically-contiguous pages. |
| 2491 | * @size: the number of bytes to allocate |
| 2492 | * @gfp_mask: GFP flags for the allocation |
| 2493 | * |
| 2494 | * This function is similar to alloc_pages(), except that it allocates the |
| 2495 | * minimum number of pages to satisfy the request. alloc_pages() can only |
| 2496 | * allocate memory in power-of-two pages. |
| 2497 | * |
| 2498 | * This function is also limited by MAX_ORDER. |
| 2499 | * |
| 2500 | * Memory allocated by this function must be released by free_pages_exact(). |
| 2501 | */ |
| 2502 | void *alloc_pages_exact(size_t size, gfp_t gfp_mask) |
| 2503 | { |
| 2504 | unsigned int order = get_order(size); |
| 2505 | unsigned long addr; |
| 2506 | |
| 2507 | addr = __get_free_pages(gfp_mask, order); |
Andi Kleen | ee85c2e | 2011-05-11 15:13:34 -0700 | [diff] [blame] | 2508 | return make_alloc_exact(addr, order, size); |
Timur Tabi | 2be0ffe | 2008-07-23 21:28:11 -0700 | [diff] [blame] | 2509 | } |
| 2510 | EXPORT_SYMBOL(alloc_pages_exact); |
| 2511 | |
| 2512 | /** |
Andi Kleen | ee85c2e | 2011-05-11 15:13:34 -0700 | [diff] [blame] | 2513 | * alloc_pages_exact_nid - allocate an exact number of physically-contiguous |
| 2514 | * pages on a node. |
Randy Dunlap | b5e6ab5 | 2011-05-16 13:16:54 -0700 | [diff] [blame] | 2515 | * @nid: the preferred node ID where memory should be allocated |
Andi Kleen | ee85c2e | 2011-05-11 15:13:34 -0700 | [diff] [blame] | 2516 | * @size: the number of bytes to allocate |
| 2517 | * @gfp_mask: GFP flags for the allocation |
| 2518 | * |
| 2519 | * Like alloc_pages_exact(), but try to allocate on node nid first before falling |
| 2520 | * back. |
| 2521 | * Note this is not alloc_pages_exact_node() which allocates on a specific node, |
| 2522 | * but is not exact. |
| 2523 | */ |
| 2524 | void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) |
| 2525 | { |
| 2526 | unsigned order = get_order(size); |
| 2527 | struct page *p = alloc_pages_node(nid, gfp_mask, order); |
| 2528 | if (!p) |
| 2529 | return NULL; |
| 2530 | return make_alloc_exact((unsigned long)page_address(p), order, size); |
| 2531 | } |
| 2532 | EXPORT_SYMBOL(alloc_pages_exact_nid); |
| 2533 | |
| 2534 | /** |
Timur Tabi | 2be0ffe | 2008-07-23 21:28:11 -0700 | [diff] [blame] | 2535 | * free_pages_exact - release memory allocated via alloc_pages_exact() |
| 2536 | * @virt: the value returned by alloc_pages_exact. |
| 2537 | * @size: size of allocation, same value as passed to alloc_pages_exact(). |
| 2538 | * |
| 2539 | * Release the memory allocated by a previous call to alloc_pages_exact. |
| 2540 | */ |
| 2541 | void free_pages_exact(void *virt, size_t size) |
| 2542 | { |
| 2543 | unsigned long addr = (unsigned long)virt; |
| 2544 | unsigned long end = addr + PAGE_ALIGN(size); |
| 2545 | |
| 2546 | while (addr < end) { |
| 2547 | free_page(addr); |
| 2548 | addr += PAGE_SIZE; |
| 2549 | } |
| 2550 | } |
| 2551 | EXPORT_SYMBOL(free_pages_exact); |
| 2552 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2553 | static unsigned int nr_free_zone_pages(int offset) |
| 2554 | { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 2555 | struct zoneref *z; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2556 | struct zone *zone; |
| 2557 | |
Martin J. Bligh | e310fd4 | 2005-07-29 22:59:18 -0700 | [diff] [blame] | 2558 | /* Just pick one node, since fallback list is circular */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2559 | unsigned int sum = 0; |
| 2560 | |
Mel Gorman | 0e88460 | 2008-04-28 02:12:14 -0700 | [diff] [blame] | 2561 | struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2562 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 2563 | for_each_zone_zonelist(zone, z, zonelist, offset) { |
Martin J. Bligh | e310fd4 | 2005-07-29 22:59:18 -0700 | [diff] [blame] | 2564 | unsigned long size = zone->present_pages; |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 2565 | unsigned long high = high_wmark_pages(zone); |
Martin J. Bligh | e310fd4 | 2005-07-29 22:59:18 -0700 | [diff] [blame] | 2566 | if (size > high) |
| 2567 | sum += size - high; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2568 | } |
| 2569 | |
| 2570 | return sum; |
| 2571 | } |
| 2572 | |
| 2573 | /* |
| 2574 | * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL |
| 2575 | */ |
| 2576 | unsigned int nr_free_buffer_pages(void) |
| 2577 | { |
Al Viro | af4ca45 | 2005-10-21 02:55:38 -0400 | [diff] [blame] | 2578 | return nr_free_zone_pages(gfp_zone(GFP_USER)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2579 | } |
Meelap Shah | c2f1a55 | 2007-07-17 04:04:39 -0700 | [diff] [blame] | 2580 | EXPORT_SYMBOL_GPL(nr_free_buffer_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2581 | |
| 2582 | /* |
| 2583 | * Amount of free RAM allocatable within all zones |
| 2584 | */ |
| 2585 | unsigned int nr_free_pagecache_pages(void) |
| 2586 | { |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 2587 | return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2588 | } |
Christoph Lameter | 08e0f6a | 2006-09-27 01:50:06 -0700 | [diff] [blame] | 2589 | |
| 2590 | static inline void show_node(struct zone *zone) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2591 | { |
Christoph Lameter | 08e0f6a | 2006-09-27 01:50:06 -0700 | [diff] [blame] | 2592 | if (NUMA_BUILD) |
Andy Whitcroft | 25ba77c | 2006-12-06 20:33:03 -0800 | [diff] [blame] | 2593 | printk("Node %d ", zone_to_nid(zone)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2594 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2595 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2596 | void si_meminfo(struct sysinfo *val) |
| 2597 | { |
| 2598 | val->totalram = totalram_pages; |
| 2599 | val->sharedram = 0; |
Christoph Lameter | d23ad42 | 2007-02-10 01:43:02 -0800 | [diff] [blame] | 2600 | val->freeram = global_page_state(NR_FREE_PAGES); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2601 | val->bufferram = nr_blockdev_pages(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2602 | val->totalhigh = totalhigh_pages; |
| 2603 | val->freehigh = nr_free_highpages(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2604 | val->mem_unit = PAGE_SIZE; |
| 2605 | } |
| 2606 | |
| 2607 | EXPORT_SYMBOL(si_meminfo); |
| 2608 | |
| 2609 | #ifdef CONFIG_NUMA |
| 2610 | void si_meminfo_node(struct sysinfo *val, int nid) |
| 2611 | { |
| 2612 | pg_data_t *pgdat = NODE_DATA(nid); |
| 2613 | |
| 2614 | val->totalram = pgdat->node_present_pages; |
Christoph Lameter | d23ad42 | 2007-02-10 01:43:02 -0800 | [diff] [blame] | 2615 | val->freeram = node_page_state(nid, NR_FREE_PAGES); |
Christoph Lameter | 98d2b0e | 2006-09-25 23:31:12 -0700 | [diff] [blame] | 2616 | #ifdef CONFIG_HIGHMEM |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2617 | val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; |
Christoph Lameter | d23ad42 | 2007-02-10 01:43:02 -0800 | [diff] [blame] | 2618 | val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM], |
| 2619 | NR_FREE_PAGES); |
Christoph Lameter | 98d2b0e | 2006-09-25 23:31:12 -0700 | [diff] [blame] | 2620 | #else |
| 2621 | val->totalhigh = 0; |
| 2622 | val->freehigh = 0; |
| 2623 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2624 | val->mem_unit = PAGE_SIZE; |
| 2625 | } |
| 2626 | #endif |
| 2627 | |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 2628 | /* |
David Rientjes | 7bf02ea | 2011-05-24 17:11:16 -0700 | [diff] [blame] | 2629 | * Determine whether the node should be displayed or not, depending on whether |
| 2630 | * SHOW_MEM_FILTER_NODES was passed to show_free_areas(). |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 2631 | */ |
David Rientjes | 7bf02ea | 2011-05-24 17:11:16 -0700 | [diff] [blame] | 2632 | bool skip_free_areas_node(unsigned int flags, int nid) |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 2633 | { |
| 2634 | bool ret = false; |
| 2635 | |
| 2636 | if (!(flags & SHOW_MEM_FILTER_NODES)) |
| 2637 | goto out; |
| 2638 | |
| 2639 | get_mems_allowed(); |
David Rientjes | 7bf02ea | 2011-05-24 17:11:16 -0700 | [diff] [blame] | 2640 | ret = !node_isset(nid, cpuset_current_mems_allowed); |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 2641 | put_mems_allowed(); |
| 2642 | out: |
| 2643 | return ret; |
| 2644 | } |
| 2645 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2646 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
| 2647 | |
| 2648 | /* |
| 2649 | * Show free area list (used inside shift_scroll-lock stuff) |
| 2650 | * We also calculate the percentage fragmentation. We do this by counting the |
| 2651 | * memory on each free list with the exception of the first item on the list. |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 2652 | * Suppresses nodes that are not allowed by current's cpuset if |
| 2653 | * SHOW_MEM_FILTER_NODES is passed. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2654 | */ |
David Rientjes | 7bf02ea | 2011-05-24 17:11:16 -0700 | [diff] [blame] | 2655 | void show_free_areas(unsigned int filter) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2656 | { |
Jes Sorensen | c724191 | 2006-09-27 01:50:05 -0700 | [diff] [blame] | 2657 | int cpu; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2658 | struct zone *zone; |
| 2659 | |
KOSAKI Motohiro | ee99c71 | 2009-03-31 15:19:31 -0700 | [diff] [blame] | 2660 | for_each_populated_zone(zone) { |
David Rientjes | 7bf02ea | 2011-05-24 17:11:16 -0700 | [diff] [blame] | 2661 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 2662 | continue; |
Jes Sorensen | c724191 | 2006-09-27 01:50:05 -0700 | [diff] [blame] | 2663 | show_node(zone); |
| 2664 | printk("%s per-cpu:\n", zone->name); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2665 | |
Dave Jones | 6b482c6 | 2005-11-10 15:45:56 -0500 | [diff] [blame] | 2666 | for_each_online_cpu(cpu) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2667 | struct per_cpu_pageset *pageset; |
| 2668 | |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 2669 | pageset = per_cpu_ptr(zone->pageset, cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2670 | |
Christoph Lameter | 3dfa572 | 2008-02-04 22:29:19 -0800 | [diff] [blame] | 2671 | printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n", |
| 2672 | cpu, pageset->pcp.high, |
| 2673 | pageset->pcp.batch, pageset->pcp.count); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2674 | } |
| 2675 | } |
| 2676 | |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 2677 | printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n" |
| 2678 | " active_file:%lu inactive_file:%lu isolated_file:%lu\n" |
Lee Schermerhorn | 7b85412 | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 2679 | " unevictable:%lu" |
Andrew Morton | b76146e | 2009-10-26 16:49:52 -0700 | [diff] [blame] | 2680 | " dirty:%lu writeback:%lu unstable:%lu\n" |
KOSAKI Motohiro | 3701b03 | 2009-09-21 17:01:29 -0700 | [diff] [blame] | 2681 | " free:%lu slab_reclaimable:%lu slab_unreclaimable:%lu\n" |
KOSAKI Motohiro | 4b02108 | 2009-09-21 17:01:33 -0700 | [diff] [blame] | 2682 | " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n", |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2683 | global_page_state(NR_ACTIVE_ANON), |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2684 | global_page_state(NR_INACTIVE_ANON), |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 2685 | global_page_state(NR_ISOLATED_ANON), |
| 2686 | global_page_state(NR_ACTIVE_FILE), |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2687 | global_page_state(NR_INACTIVE_FILE), |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 2688 | global_page_state(NR_ISOLATED_FILE), |
Lee Schermerhorn | 7b85412 | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 2689 | global_page_state(NR_UNEVICTABLE), |
Christoph Lameter | b1e7a8f | 2006-06-30 01:55:39 -0700 | [diff] [blame] | 2690 | global_page_state(NR_FILE_DIRTY), |
Christoph Lameter | ce866b3 | 2006-06-30 01:55:40 -0700 | [diff] [blame] | 2691 | global_page_state(NR_WRITEBACK), |
Christoph Lameter | fd39fc8 | 2006-06-30 01:55:40 -0700 | [diff] [blame] | 2692 | global_page_state(NR_UNSTABLE_NFS), |
Christoph Lameter | d23ad42 | 2007-02-10 01:43:02 -0800 | [diff] [blame] | 2693 | global_page_state(NR_FREE_PAGES), |
KOSAKI Motohiro | 3701b03 | 2009-09-21 17:01:29 -0700 | [diff] [blame] | 2694 | global_page_state(NR_SLAB_RECLAIMABLE), |
| 2695 | global_page_state(NR_SLAB_UNRECLAIMABLE), |
Christoph Lameter | 65ba55f | 2006-06-30 01:55:34 -0700 | [diff] [blame] | 2696 | global_page_state(NR_FILE_MAPPED), |
KOSAKI Motohiro | 4b02108 | 2009-09-21 17:01:33 -0700 | [diff] [blame] | 2697 | global_page_state(NR_SHMEM), |
Andrew Morton | a25700a | 2007-02-08 14:20:40 -0800 | [diff] [blame] | 2698 | global_page_state(NR_PAGETABLE), |
| 2699 | global_page_state(NR_BOUNCE)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2700 | |
KOSAKI Motohiro | ee99c71 | 2009-03-31 15:19:31 -0700 | [diff] [blame] | 2701 | for_each_populated_zone(zone) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2702 | int i; |
| 2703 | |
David Rientjes | 7bf02ea | 2011-05-24 17:11:16 -0700 | [diff] [blame] | 2704 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 2705 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2706 | show_node(zone); |
| 2707 | printk("%s" |
| 2708 | " free:%lukB" |
| 2709 | " min:%lukB" |
| 2710 | " low:%lukB" |
| 2711 | " high:%lukB" |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2712 | " active_anon:%lukB" |
| 2713 | " inactive_anon:%lukB" |
| 2714 | " active_file:%lukB" |
| 2715 | " inactive_file:%lukB" |
Lee Schermerhorn | 7b85412 | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 2716 | " unevictable:%lukB" |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 2717 | " isolated(anon):%lukB" |
| 2718 | " isolated(file):%lukB" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2719 | " present:%lukB" |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 2720 | " mlocked:%lukB" |
| 2721 | " dirty:%lukB" |
| 2722 | " writeback:%lukB" |
| 2723 | " mapped:%lukB" |
KOSAKI Motohiro | 4b02108 | 2009-09-21 17:01:33 -0700 | [diff] [blame] | 2724 | " shmem:%lukB" |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 2725 | " slab_reclaimable:%lukB" |
| 2726 | " slab_unreclaimable:%lukB" |
KOSAKI Motohiro | c6a7f57 | 2009-09-21 17:01:32 -0700 | [diff] [blame] | 2727 | " kernel_stack:%lukB" |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 2728 | " pagetables:%lukB" |
| 2729 | " unstable:%lukB" |
| 2730 | " bounce:%lukB" |
| 2731 | " writeback_tmp:%lukB" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2732 | " pages_scanned:%lu" |
| 2733 | " all_unreclaimable? %s" |
| 2734 | "\n", |
| 2735 | zone->name, |
Mel Gorman | 88f5acf | 2011-01-13 15:45:41 -0800 | [diff] [blame] | 2736 | K(zone_page_state(zone, NR_FREE_PAGES)), |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 2737 | K(min_wmark_pages(zone)), |
| 2738 | K(low_wmark_pages(zone)), |
| 2739 | K(high_wmark_pages(zone)), |
Rik van Riel | 4f98a2f | 2008-10-18 20:26:32 -0700 | [diff] [blame] | 2740 | K(zone_page_state(zone, NR_ACTIVE_ANON)), |
| 2741 | K(zone_page_state(zone, NR_INACTIVE_ANON)), |
| 2742 | K(zone_page_state(zone, NR_ACTIVE_FILE)), |
| 2743 | K(zone_page_state(zone, NR_INACTIVE_FILE)), |
Lee Schermerhorn | 7b85412 | 2008-10-18 20:26:40 -0700 | [diff] [blame] | 2744 | K(zone_page_state(zone, NR_UNEVICTABLE)), |
KOSAKI Motohiro | a731286 | 2009-09-21 17:01:37 -0700 | [diff] [blame] | 2745 | K(zone_page_state(zone, NR_ISOLATED_ANON)), |
| 2746 | K(zone_page_state(zone, NR_ISOLATED_FILE)), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2747 | K(zone->present_pages), |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 2748 | K(zone_page_state(zone, NR_MLOCK)), |
| 2749 | K(zone_page_state(zone, NR_FILE_DIRTY)), |
| 2750 | K(zone_page_state(zone, NR_WRITEBACK)), |
| 2751 | K(zone_page_state(zone, NR_FILE_MAPPED)), |
KOSAKI Motohiro | 4b02108 | 2009-09-21 17:01:33 -0700 | [diff] [blame] | 2752 | K(zone_page_state(zone, NR_SHMEM)), |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 2753 | K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)), |
| 2754 | K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)), |
KOSAKI Motohiro | c6a7f57 | 2009-09-21 17:01:32 -0700 | [diff] [blame] | 2755 | zone_page_state(zone, NR_KERNEL_STACK) * |
| 2756 | THREAD_SIZE / 1024, |
KOSAKI Motohiro | 4a0aa73 | 2009-09-21 17:01:30 -0700 | [diff] [blame] | 2757 | K(zone_page_state(zone, NR_PAGETABLE)), |
| 2758 | K(zone_page_state(zone, NR_UNSTABLE_NFS)), |
| 2759 | K(zone_page_state(zone, NR_BOUNCE)), |
| 2760 | K(zone_page_state(zone, NR_WRITEBACK_TEMP)), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2761 | zone->pages_scanned, |
KOSAKI Motohiro | 93e4a89 | 2010-03-05 13:41:55 -0800 | [diff] [blame] | 2762 | (zone->all_unreclaimable ? "yes" : "no") |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2763 | ); |
| 2764 | printk("lowmem_reserve[]:"); |
| 2765 | for (i = 0; i < MAX_NR_ZONES; i++) |
| 2766 | printk(" %lu", zone->lowmem_reserve[i]); |
| 2767 | printk("\n"); |
| 2768 | } |
| 2769 | |
KOSAKI Motohiro | ee99c71 | 2009-03-31 15:19:31 -0700 | [diff] [blame] | 2770 | for_each_populated_zone(zone) { |
Kirill Korotaev | 8f9de51 | 2006-06-23 02:03:50 -0700 | [diff] [blame] | 2771 | unsigned long nr[MAX_ORDER], flags, order, total = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2772 | |
David Rientjes | 7bf02ea | 2011-05-24 17:11:16 -0700 | [diff] [blame] | 2773 | if (skip_free_areas_node(filter, zone_to_nid(zone))) |
David Rientjes | ddd588b | 2011-03-22 16:30:46 -0700 | [diff] [blame] | 2774 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2775 | show_node(zone); |
| 2776 | printk("%s: ", zone->name); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2777 | |
| 2778 | spin_lock_irqsave(&zone->lock, flags); |
| 2779 | for (order = 0; order < MAX_ORDER; order++) { |
Kirill Korotaev | 8f9de51 | 2006-06-23 02:03:50 -0700 | [diff] [blame] | 2780 | nr[order] = zone->free_area[order].nr_free; |
| 2781 | total += nr[order] << order; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2782 | } |
| 2783 | spin_unlock_irqrestore(&zone->lock, flags); |
Kirill Korotaev | 8f9de51 | 2006-06-23 02:03:50 -0700 | [diff] [blame] | 2784 | for (order = 0; order < MAX_ORDER; order++) |
| 2785 | printk("%lu*%lukB ", nr[order], K(1UL) << order); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2786 | printk("= %lukB\n", K(total)); |
| 2787 | } |
| 2788 | |
Larry Woodman | e6f3602 | 2008-02-04 22:29:30 -0800 | [diff] [blame] | 2789 | printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES)); |
| 2790 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2791 | show_swap_cache_info(); |
| 2792 | } |
| 2793 | |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 2794 | static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref) |
| 2795 | { |
| 2796 | zoneref->zone = zone; |
| 2797 | zoneref->zone_idx = zone_idx(zone); |
| 2798 | } |
| 2799 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2800 | /* |
| 2801 | * Builds allocation fallback zone lists. |
Christoph Lameter | 1a93205 | 2006-01-06 00:11:16 -0800 | [diff] [blame] | 2802 | * |
| 2803 | * Add all populated zones of a node to the zonelist. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2804 | */ |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2805 | static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist, |
| 2806 | int nr_zones, enum zone_type zone_type) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2807 | { |
Christoph Lameter | 1a93205 | 2006-01-06 00:11:16 -0800 | [diff] [blame] | 2808 | struct zone *zone; |
| 2809 | |
Christoph Lameter | 98d2b0e | 2006-09-25 23:31:12 -0700 | [diff] [blame] | 2810 | BUG_ON(zone_type >= MAX_NR_ZONES); |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 2811 | zone_type++; |
Christoph Lameter | 02a68a5 | 2006-01-06 00:11:18 -0800 | [diff] [blame] | 2812 | |
| 2813 | do { |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 2814 | zone_type--; |
Christoph Lameter | 070f803 | 2006-01-06 00:11:19 -0800 | [diff] [blame] | 2815 | zone = pgdat->node_zones + zone_type; |
Christoph Lameter | 1a93205 | 2006-01-06 00:11:16 -0800 | [diff] [blame] | 2816 | if (populated_zone(zone)) { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 2817 | zoneref_set_zone(zone, |
| 2818 | &zonelist->_zonerefs[nr_zones++]); |
Christoph Lameter | 070f803 | 2006-01-06 00:11:19 -0800 | [diff] [blame] | 2819 | check_highest_zone(zone_type); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2820 | } |
Christoph Lameter | 02a68a5 | 2006-01-06 00:11:18 -0800 | [diff] [blame] | 2821 | |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 2822 | } while (zone_type); |
Christoph Lameter | 070f803 | 2006-01-06 00:11:19 -0800 | [diff] [blame] | 2823 | return nr_zones; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2824 | } |
| 2825 | |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2826 | |
| 2827 | /* |
| 2828 | * zonelist_order: |
| 2829 | * 0 = automatic detection of better ordering. |
| 2830 | * 1 = order by ([node] distance, -zonetype) |
| 2831 | * 2 = order by (-zonetype, [node] distance) |
| 2832 | * |
| 2833 | * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create |
| 2834 | * the same zonelist. So only NUMA can configure this param. |
| 2835 | */ |
| 2836 | #define ZONELIST_ORDER_DEFAULT 0 |
| 2837 | #define ZONELIST_ORDER_NODE 1 |
| 2838 | #define ZONELIST_ORDER_ZONE 2 |
| 2839 | |
| 2840 | /* zonelist order in the kernel. |
| 2841 | * set_zonelist_order() will set this to NODE or ZONE. |
| 2842 | */ |
| 2843 | static int current_zonelist_order = ZONELIST_ORDER_DEFAULT; |
| 2844 | static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"}; |
| 2845 | |
| 2846 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2847 | #ifdef CONFIG_NUMA |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2848 | /* The value user specified ....changed by config */ |
| 2849 | static int user_zonelist_order = ZONELIST_ORDER_DEFAULT; |
| 2850 | /* string for sysctl */ |
| 2851 | #define NUMA_ZONELIST_ORDER_LEN 16 |
| 2852 | char numa_zonelist_order[16] = "default"; |
| 2853 | |
| 2854 | /* |
| 2855 | * interface for configure zonelist ordering. |
| 2856 | * command line option "numa_zonelist_order" |
| 2857 | * = "[dD]efault - default, automatic configuration. |
| 2858 | * = "[nN]ode - order by node locality, then by zone within node |
| 2859 | * = "[zZ]one - order by zone, then by locality within zone |
| 2860 | */ |
| 2861 | |
| 2862 | static int __parse_numa_zonelist_order(char *s) |
| 2863 | { |
| 2864 | if (*s == 'd' || *s == 'D') { |
| 2865 | user_zonelist_order = ZONELIST_ORDER_DEFAULT; |
| 2866 | } else if (*s == 'n' || *s == 'N') { |
| 2867 | user_zonelist_order = ZONELIST_ORDER_NODE; |
| 2868 | } else if (*s == 'z' || *s == 'Z') { |
| 2869 | user_zonelist_order = ZONELIST_ORDER_ZONE; |
| 2870 | } else { |
| 2871 | printk(KERN_WARNING |
| 2872 | "Ignoring invalid numa_zonelist_order value: " |
| 2873 | "%s\n", s); |
| 2874 | return -EINVAL; |
| 2875 | } |
| 2876 | return 0; |
| 2877 | } |
| 2878 | |
| 2879 | static __init int setup_numa_zonelist_order(char *s) |
| 2880 | { |
Volodymyr G. Lukiianyk | ecb256f | 2011-01-13 15:46:26 -0800 | [diff] [blame] | 2881 | int ret; |
| 2882 | |
| 2883 | if (!s) |
| 2884 | return 0; |
| 2885 | |
| 2886 | ret = __parse_numa_zonelist_order(s); |
| 2887 | if (ret == 0) |
| 2888 | strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN); |
| 2889 | |
| 2890 | return ret; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2891 | } |
| 2892 | early_param("numa_zonelist_order", setup_numa_zonelist_order); |
| 2893 | |
| 2894 | /* |
| 2895 | * sysctl handler for numa_zonelist_order |
| 2896 | */ |
| 2897 | int numa_zonelist_order_handler(ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 2898 | void __user *buffer, size_t *length, |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2899 | loff_t *ppos) |
| 2900 | { |
| 2901 | char saved_string[NUMA_ZONELIST_ORDER_LEN]; |
| 2902 | int ret; |
Andi Kleen | 443c6f1 | 2009-12-23 21:00:47 +0100 | [diff] [blame] | 2903 | static DEFINE_MUTEX(zl_order_mutex); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2904 | |
Andi Kleen | 443c6f1 | 2009-12-23 21:00:47 +0100 | [diff] [blame] | 2905 | mutex_lock(&zl_order_mutex); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2906 | if (write) |
Andi Kleen | 443c6f1 | 2009-12-23 21:00:47 +0100 | [diff] [blame] | 2907 | strcpy(saved_string, (char*)table->data); |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 2908 | ret = proc_dostring(table, write, buffer, length, ppos); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2909 | if (ret) |
Andi Kleen | 443c6f1 | 2009-12-23 21:00:47 +0100 | [diff] [blame] | 2910 | goto out; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2911 | if (write) { |
| 2912 | int oldval = user_zonelist_order; |
| 2913 | if (__parse_numa_zonelist_order((char*)table->data)) { |
| 2914 | /* |
| 2915 | * bogus value. restore saved string |
| 2916 | */ |
| 2917 | strncpy((char*)table->data, saved_string, |
| 2918 | NUMA_ZONELIST_ORDER_LEN); |
| 2919 | user_zonelist_order = oldval; |
Haicheng Li | 4eaf3f6 | 2010-05-24 14:32:52 -0700 | [diff] [blame] | 2920 | } else if (oldval != user_zonelist_order) { |
| 2921 | mutex_lock(&zonelists_mutex); |
Haicheng Li | 1f52250 | 2010-05-24 14:32:51 -0700 | [diff] [blame] | 2922 | build_all_zonelists(NULL); |
Haicheng Li | 4eaf3f6 | 2010-05-24 14:32:52 -0700 | [diff] [blame] | 2923 | mutex_unlock(&zonelists_mutex); |
| 2924 | } |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2925 | } |
Andi Kleen | 443c6f1 | 2009-12-23 21:00:47 +0100 | [diff] [blame] | 2926 | out: |
| 2927 | mutex_unlock(&zl_order_mutex); |
| 2928 | return ret; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2929 | } |
| 2930 | |
| 2931 | |
Christoph Lameter | 62bc62a | 2009-06-16 15:32:15 -0700 | [diff] [blame] | 2932 | #define MAX_NODE_LOAD (nr_online_nodes) |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2933 | static int node_load[MAX_NUMNODES]; |
| 2934 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2935 | /** |
Pavel Pisa | 4dc3b16 | 2005-05-01 08:59:25 -0700 | [diff] [blame] | 2936 | * find_next_best_node - find the next node that should appear in a given node's fallback list |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2937 | * @node: node whose fallback list we're appending |
| 2938 | * @used_node_mask: nodemask_t of already used nodes |
| 2939 | * |
| 2940 | * We use a number of factors to determine which is the next node that should |
| 2941 | * appear on a given node's fallback list. The node should not have appeared |
| 2942 | * already in @node's fallback list, and it should be the next closest node |
| 2943 | * according to the distance array (which contains arbitrary distance values |
| 2944 | * from each node to each node in the system), and should also prefer nodes |
| 2945 | * with no CPUs, since presumably they'll have very little allocation pressure |
| 2946 | * on them otherwise. |
| 2947 | * It returns -1 if no node is found. |
| 2948 | */ |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2949 | static int find_next_best_node(int node, nodemask_t *used_node_mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2950 | { |
Linus Torvalds | 4cf808e | 2006-02-17 20:38:21 +0100 | [diff] [blame] | 2951 | int n, val; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2952 | int min_val = INT_MAX; |
| 2953 | int best_node = -1; |
Rusty Russell | a70f730 | 2009-03-13 14:49:46 +1030 | [diff] [blame] | 2954 | const struct cpumask *tmp = cpumask_of_node(0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2955 | |
Linus Torvalds | 4cf808e | 2006-02-17 20:38:21 +0100 | [diff] [blame] | 2956 | /* Use the local node if we haven't already */ |
| 2957 | if (!node_isset(node, *used_node_mask)) { |
| 2958 | node_set(node, *used_node_mask); |
| 2959 | return node; |
| 2960 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2961 | |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 2962 | for_each_node_state(n, N_HIGH_MEMORY) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2963 | |
| 2964 | /* Don't want a node to appear more than once */ |
| 2965 | if (node_isset(n, *used_node_mask)) |
| 2966 | continue; |
| 2967 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2968 | /* Use the distance array to find the distance */ |
| 2969 | val = node_distance(node, n); |
| 2970 | |
Linus Torvalds | 4cf808e | 2006-02-17 20:38:21 +0100 | [diff] [blame] | 2971 | /* Penalize nodes under us ("prefer the next node") */ |
| 2972 | val += (n < node); |
| 2973 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2974 | /* Give preference to headless and unused nodes */ |
Rusty Russell | a70f730 | 2009-03-13 14:49:46 +1030 | [diff] [blame] | 2975 | tmp = cpumask_of_node(n); |
| 2976 | if (!cpumask_empty(tmp)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2977 | val += PENALTY_FOR_NODE_WITH_CPUS; |
| 2978 | |
| 2979 | /* Slight preference for less loaded node */ |
| 2980 | val *= (MAX_NODE_LOAD*MAX_NUMNODES); |
| 2981 | val += node_load[n]; |
| 2982 | |
| 2983 | if (val < min_val) { |
| 2984 | min_val = val; |
| 2985 | best_node = n; |
| 2986 | } |
| 2987 | } |
| 2988 | |
| 2989 | if (best_node >= 0) |
| 2990 | node_set(best_node, *used_node_mask); |
| 2991 | |
| 2992 | return best_node; |
| 2993 | } |
| 2994 | |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 2995 | |
| 2996 | /* |
| 2997 | * Build zonelists ordered by node and zones within node. |
| 2998 | * This results in maximum locality--normal zone overflows into local |
| 2999 | * DMA zone, if any--but risks exhausting DMA zone. |
| 3000 | */ |
| 3001 | static void build_zonelists_in_node_order(pg_data_t *pgdat, int node) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3002 | { |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3003 | int j; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3004 | struct zonelist *zonelist; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3005 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 3006 | zonelist = &pgdat->node_zonelists[0]; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 3007 | for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++) |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 3008 | ; |
| 3009 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, |
| 3010 | MAX_NR_ZONES - 1); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 3011 | zonelist->_zonerefs[j].zone = NULL; |
| 3012 | zonelist->_zonerefs[j].zone_idx = 0; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3013 | } |
| 3014 | |
| 3015 | /* |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 3016 | * Build gfp_thisnode zonelists |
| 3017 | */ |
| 3018 | static void build_thisnode_zonelists(pg_data_t *pgdat) |
| 3019 | { |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 3020 | int j; |
| 3021 | struct zonelist *zonelist; |
| 3022 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 3023 | zonelist = &pgdat->node_zonelists[1]; |
| 3024 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 3025 | zonelist->_zonerefs[j].zone = NULL; |
| 3026 | zonelist->_zonerefs[j].zone_idx = 0; |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 3027 | } |
| 3028 | |
| 3029 | /* |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3030 | * Build zonelists ordered by zone and nodes within zones. |
| 3031 | * This results in conserving DMA zone[s] until all Normal memory is |
| 3032 | * exhausted, but results in overflowing to remote node while memory |
| 3033 | * may still exist in local DMA zone. |
| 3034 | */ |
| 3035 | static int node_order[MAX_NUMNODES]; |
| 3036 | |
| 3037 | static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes) |
| 3038 | { |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3039 | int pos, j, node; |
| 3040 | int zone_type; /* needs to be signed */ |
| 3041 | struct zone *z; |
| 3042 | struct zonelist *zonelist; |
| 3043 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 3044 | zonelist = &pgdat->node_zonelists[0]; |
| 3045 | pos = 0; |
| 3046 | for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) { |
| 3047 | for (j = 0; j < nr_nodes; j++) { |
| 3048 | node = node_order[j]; |
| 3049 | z = &NODE_DATA(node)->node_zones[zone_type]; |
| 3050 | if (populated_zone(z)) { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 3051 | zoneref_set_zone(z, |
| 3052 | &zonelist->_zonerefs[pos++]); |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 3053 | check_highest_zone(zone_type); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3054 | } |
| 3055 | } |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3056 | } |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 3057 | zonelist->_zonerefs[pos].zone = NULL; |
| 3058 | zonelist->_zonerefs[pos].zone_idx = 0; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3059 | } |
| 3060 | |
| 3061 | static int default_zonelist_order(void) |
| 3062 | { |
| 3063 | int nid, zone_type; |
| 3064 | unsigned long low_kmem_size,total_size; |
| 3065 | struct zone *z; |
| 3066 | int average_size; |
| 3067 | /* |
Thomas Weber | 8839316 | 2010-03-16 11:47:56 +0100 | [diff] [blame] | 3068 | * ZONE_DMA and ZONE_DMA32 can be very small area in the system. |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3069 | * If they are really small and used heavily, the system can fall |
| 3070 | * into OOM very easily. |
David Rientjes | e325c90 | 2010-05-24 14:32:13 -0700 | [diff] [blame] | 3071 | * This function detect ZONE_DMA/DMA32 size and configures zone order. |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3072 | */ |
| 3073 | /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */ |
| 3074 | low_kmem_size = 0; |
| 3075 | total_size = 0; |
| 3076 | for_each_online_node(nid) { |
| 3077 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
| 3078 | z = &NODE_DATA(nid)->node_zones[zone_type]; |
| 3079 | if (populated_zone(z)) { |
| 3080 | if (zone_type < ZONE_NORMAL) |
| 3081 | low_kmem_size += z->present_pages; |
| 3082 | total_size += z->present_pages; |
David Rientjes | e325c90 | 2010-05-24 14:32:13 -0700 | [diff] [blame] | 3083 | } else if (zone_type == ZONE_NORMAL) { |
| 3084 | /* |
| 3085 | * If any node has only lowmem, then node order |
| 3086 | * is preferred to allow kernel allocations |
| 3087 | * locally; otherwise, they can easily infringe |
| 3088 | * on other nodes when there is an abundance of |
| 3089 | * lowmem available to allocate from. |
| 3090 | */ |
| 3091 | return ZONELIST_ORDER_NODE; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3092 | } |
| 3093 | } |
| 3094 | } |
| 3095 | if (!low_kmem_size || /* there are no DMA area. */ |
| 3096 | low_kmem_size > total_size/2) /* DMA/DMA32 is big. */ |
| 3097 | return ZONELIST_ORDER_NODE; |
| 3098 | /* |
| 3099 | * look into each node's config. |
| 3100 | * If there is a node whose DMA/DMA32 memory is very big area on |
| 3101 | * local memory, NODE_ORDER may be suitable. |
| 3102 | */ |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 3103 | average_size = total_size / |
| 3104 | (nodes_weight(node_states[N_HIGH_MEMORY]) + 1); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3105 | for_each_online_node(nid) { |
| 3106 | low_kmem_size = 0; |
| 3107 | total_size = 0; |
| 3108 | for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) { |
| 3109 | z = &NODE_DATA(nid)->node_zones[zone_type]; |
| 3110 | if (populated_zone(z)) { |
| 3111 | if (zone_type < ZONE_NORMAL) |
| 3112 | low_kmem_size += z->present_pages; |
| 3113 | total_size += z->present_pages; |
| 3114 | } |
| 3115 | } |
| 3116 | if (low_kmem_size && |
| 3117 | total_size > average_size && /* ignore small node */ |
| 3118 | low_kmem_size > total_size * 70/100) |
| 3119 | return ZONELIST_ORDER_NODE; |
| 3120 | } |
| 3121 | return ZONELIST_ORDER_ZONE; |
| 3122 | } |
| 3123 | |
| 3124 | static void set_zonelist_order(void) |
| 3125 | { |
| 3126 | if (user_zonelist_order == ZONELIST_ORDER_DEFAULT) |
| 3127 | current_zonelist_order = default_zonelist_order(); |
| 3128 | else |
| 3129 | current_zonelist_order = user_zonelist_order; |
| 3130 | } |
| 3131 | |
| 3132 | static void build_zonelists(pg_data_t *pgdat) |
| 3133 | { |
| 3134 | int j, node, load; |
| 3135 | enum zone_type i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3136 | nodemask_t used_mask; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3137 | int local_node, prev_node; |
| 3138 | struct zonelist *zonelist; |
| 3139 | int order = current_zonelist_order; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3140 | |
| 3141 | /* initialize zonelists */ |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 3142 | for (i = 0; i < MAX_ZONELISTS; i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3143 | zonelist = pgdat->node_zonelists + i; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 3144 | zonelist->_zonerefs[0].zone = NULL; |
| 3145 | zonelist->_zonerefs[0].zone_idx = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3146 | } |
| 3147 | |
| 3148 | /* NUMA-aware ordering of nodes */ |
| 3149 | local_node = pgdat->node_id; |
Christoph Lameter | 62bc62a | 2009-06-16 15:32:15 -0700 | [diff] [blame] | 3150 | load = nr_online_nodes; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3151 | prev_node = local_node; |
| 3152 | nodes_clear(used_mask); |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3153 | |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3154 | memset(node_order, 0, sizeof(node_order)); |
| 3155 | j = 0; |
| 3156 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3157 | while ((node = find_next_best_node(local_node, &used_mask)) >= 0) { |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3158 | int distance = node_distance(local_node, node); |
| 3159 | |
| 3160 | /* |
| 3161 | * If another node is sufficiently far away then it is better |
| 3162 | * to reclaim pages in a zone before going off node. |
| 3163 | */ |
| 3164 | if (distance > RECLAIM_DISTANCE) |
| 3165 | zone_reclaim_mode = 1; |
| 3166 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3167 | /* |
| 3168 | * We don't want to pressure a particular node. |
| 3169 | * So adding penalty to the first node in same |
| 3170 | * distance group to make it round-robin. |
| 3171 | */ |
Christoph Lameter | 9eeff23 | 2006-01-18 17:42:31 -0800 | [diff] [blame] | 3172 | if (distance != node_distance(local_node, prev_node)) |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3173 | node_load[node] = load; |
| 3174 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3175 | prev_node = node; |
| 3176 | load--; |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3177 | if (order == ZONELIST_ORDER_NODE) |
| 3178 | build_zonelists_in_node_order(pgdat, node); |
| 3179 | else |
| 3180 | node_order[j++] = node; /* remember order */ |
| 3181 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3182 | |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3183 | if (order == ZONELIST_ORDER_ZONE) { |
| 3184 | /* calculate node order -- i.e., DMA last! */ |
| 3185 | build_zonelists_in_zone_order(pgdat, j); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3186 | } |
Christoph Lameter | 523b945 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 3187 | |
| 3188 | build_thisnode_zonelists(pgdat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3189 | } |
| 3190 | |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 3191 | /* Construct the zonelist performance cache - see further mmzone.h */ |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3192 | static void build_zonelist_cache(pg_data_t *pgdat) |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 3193 | { |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 3194 | struct zonelist *zonelist; |
| 3195 | struct zonelist_cache *zlc; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 3196 | struct zoneref *z; |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 3197 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 3198 | zonelist = &pgdat->node_zonelists[0]; |
| 3199 | zonelist->zlcache_ptr = zlc = &zonelist->zlcache; |
| 3200 | bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 3201 | for (z = zonelist->_zonerefs; z->zone; z++) |
| 3202 | zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z); |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 3203 | } |
| 3204 | |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 3205 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
| 3206 | /* |
| 3207 | * Return node id of node used for "local" allocations. |
| 3208 | * I.e., first node id of first zone in arg node's generic zonelist. |
| 3209 | * Used for initializing percpu 'numa_mem', which is used primarily |
| 3210 | * for kernel allocations, so use GFP_KERNEL flags to locate zonelist. |
| 3211 | */ |
| 3212 | int local_memory_node(int node) |
| 3213 | { |
| 3214 | struct zone *zone; |
| 3215 | |
| 3216 | (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL), |
| 3217 | gfp_zone(GFP_KERNEL), |
| 3218 | NULL, |
| 3219 | &zone); |
| 3220 | return zone->node; |
| 3221 | } |
| 3222 | #endif |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3223 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3224 | #else /* CONFIG_NUMA */ |
| 3225 | |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3226 | static void set_zonelist_order(void) |
| 3227 | { |
| 3228 | current_zonelist_order = ZONELIST_ORDER_ZONE; |
| 3229 | } |
| 3230 | |
| 3231 | static void build_zonelists(pg_data_t *pgdat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3232 | { |
Christoph Lameter | 19655d3 | 2006-09-25 23:31:19 -0700 | [diff] [blame] | 3233 | int node, local_node; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 3234 | enum zone_type j; |
| 3235 | struct zonelist *zonelist; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3236 | |
| 3237 | local_node = pgdat->node_id; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3238 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 3239 | zonelist = &pgdat->node_zonelists[0]; |
| 3240 | j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3241 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 3242 | /* |
| 3243 | * Now we build the zonelist so that it contains the zones |
| 3244 | * of all the other nodes. |
| 3245 | * We don't want to pressure a particular node, so when |
| 3246 | * building the zones for node N, we make sure that the |
| 3247 | * zones coming right after the local ones are those from |
| 3248 | * node N+1 (modulo N) |
| 3249 | */ |
| 3250 | for (node = local_node + 1; node < MAX_NUMNODES; node++) { |
| 3251 | if (!node_online(node)) |
| 3252 | continue; |
| 3253 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, |
| 3254 | MAX_NR_ZONES - 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3255 | } |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 3256 | for (node = 0; node < local_node; node++) { |
| 3257 | if (!node_online(node)) |
| 3258 | continue; |
| 3259 | j = build_zonelists_node(NODE_DATA(node), zonelist, j, |
| 3260 | MAX_NR_ZONES - 1); |
| 3261 | } |
| 3262 | |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 3263 | zonelist->_zonerefs[j].zone = NULL; |
| 3264 | zonelist->_zonerefs[j].zone_idx = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3265 | } |
| 3266 | |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 3267 | /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */ |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3268 | static void build_zonelist_cache(pg_data_t *pgdat) |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 3269 | { |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 3270 | pgdat->node_zonelists[0].zlcache_ptr = NULL; |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 3271 | } |
| 3272 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3273 | #endif /* CONFIG_NUMA */ |
| 3274 | |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 3275 | /* |
| 3276 | * Boot pageset table. One per cpu which is going to be used for all |
| 3277 | * zones and all nodes. The parameters will be set in such a way |
| 3278 | * that an item put on a list will immediately be handed over to |
| 3279 | * the buddy list. This is safe since pageset manipulation is done |
| 3280 | * with interrupts disabled. |
| 3281 | * |
| 3282 | * The boot_pagesets must be kept even after bootup is complete for |
| 3283 | * unused processors and/or zones. They do play a role for bootstrapping |
| 3284 | * hotplugged processors. |
| 3285 | * |
| 3286 | * zoneinfo_show() and maybe other functions do |
| 3287 | * not check if the processor is online before following the pageset pointer. |
| 3288 | * Other parts of the kernel may not check if the zone is available. |
| 3289 | */ |
| 3290 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch); |
| 3291 | static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset); |
Haicheng Li | 1f52250 | 2010-05-24 14:32:51 -0700 | [diff] [blame] | 3292 | static void setup_zone_pageset(struct zone *zone); |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 3293 | |
Haicheng Li | 4eaf3f6 | 2010-05-24 14:32:52 -0700 | [diff] [blame] | 3294 | /* |
| 3295 | * Global mutex to protect against size modification of zonelists |
| 3296 | * as well as to serialize pageset setup for the new populated zone. |
| 3297 | */ |
| 3298 | DEFINE_MUTEX(zonelists_mutex); |
| 3299 | |
Rusty Russell | 9b1a4d3 | 2008-07-28 12:16:30 -0500 | [diff] [blame] | 3300 | /* return values int ....just for stop_machine() */ |
Haicheng Li | 1f52250 | 2010-05-24 14:32:51 -0700 | [diff] [blame] | 3301 | static __init_refok int __build_all_zonelists(void *data) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3302 | { |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 3303 | int nid; |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 3304 | int cpu; |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 3305 | |
Bo Liu | 7f9cfb3 | 2009-08-18 14:11:19 -0700 | [diff] [blame] | 3306 | #ifdef CONFIG_NUMA |
| 3307 | memset(node_load, 0, sizeof(node_load)); |
| 3308 | #endif |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 3309 | for_each_online_node(nid) { |
Christoph Lameter | 7ea1530 | 2007-10-16 01:25:29 -0700 | [diff] [blame] | 3310 | pg_data_t *pgdat = NODE_DATA(nid); |
| 3311 | |
| 3312 | build_zonelists(pgdat); |
| 3313 | build_zonelist_cache(pgdat); |
Paul Jackson | 9276b1bc | 2006-12-06 20:31:48 -0800 | [diff] [blame] | 3314 | } |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 3315 | |
| 3316 | /* |
| 3317 | * Initialize the boot_pagesets that are going to be used |
| 3318 | * for bootstrapping processors. The real pagesets for |
| 3319 | * each zone will be allocated later when the per cpu |
| 3320 | * allocator is available. |
| 3321 | * |
| 3322 | * boot_pagesets are used also for bootstrapping offline |
| 3323 | * cpus if the system is already booted because the pagesets |
| 3324 | * are needed to initialize allocators on a specific cpu too. |
| 3325 | * F.e. the percpu allocator needs the page allocator which |
| 3326 | * needs the percpu allocator in order to allocate its pagesets |
| 3327 | * (a chicken-egg dilemma). |
| 3328 | */ |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 3329 | for_each_possible_cpu(cpu) { |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 3330 | setup_pageset(&per_cpu(boot_pageset, cpu), 0); |
| 3331 | |
Lee Schermerhorn | 7aac789 | 2010-05-26 14:45:00 -0700 | [diff] [blame] | 3332 | #ifdef CONFIG_HAVE_MEMORYLESS_NODES |
| 3333 | /* |
| 3334 | * We now know the "local memory node" for each node-- |
| 3335 | * i.e., the node of the first zone in the generic zonelist. |
| 3336 | * Set up numa_mem percpu variable for on-line cpus. During |
| 3337 | * boot, only the boot cpu should be on-line; we'll init the |
| 3338 | * secondary cpus' numa_mem as they come on-line. During |
| 3339 | * node/memory hotplug, we'll fixup all on-line cpus. |
| 3340 | */ |
| 3341 | if (cpu_online(cpu)) |
| 3342 | set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu))); |
| 3343 | #endif |
| 3344 | } |
| 3345 | |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 3346 | return 0; |
| 3347 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3348 | |
Haicheng Li | 4eaf3f6 | 2010-05-24 14:32:52 -0700 | [diff] [blame] | 3349 | /* |
| 3350 | * Called with zonelists_mutex held always |
| 3351 | * unless system_state == SYSTEM_BOOTING. |
| 3352 | */ |
Paul Mundt | 9f6ae44 | 2011-04-14 15:21:57 -0700 | [diff] [blame] | 3353 | void __ref build_all_zonelists(void *data) |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 3354 | { |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3355 | set_zonelist_order(); |
| 3356 | |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 3357 | if (system_state == SYSTEM_BOOTING) { |
Randy Dunlap | 423b41d | 2006-09-27 01:50:12 -0700 | [diff] [blame] | 3358 | __build_all_zonelists(NULL); |
Mel Gorman | 68ad8df | 2008-07-23 21:26:52 -0700 | [diff] [blame] | 3359 | mminit_verify_zonelist(); |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 3360 | cpuset_init_current_mems_allowed(); |
| 3361 | } else { |
Simon Arlott | 183ff22 | 2007-10-20 01:27:18 +0200 | [diff] [blame] | 3362 | /* we have to stop all cpus to guarantee there is no user |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 3363 | of zonelist */ |
KAMEZAWA Hiroyuki | e9959f0 | 2010-11-24 12:57:09 -0800 | [diff] [blame] | 3364 | #ifdef CONFIG_MEMORY_HOTPLUG |
| 3365 | if (data) |
| 3366 | setup_zone_pageset((struct zone *)data); |
| 3367 | #endif |
| 3368 | stop_machine(__build_all_zonelists, NULL, NULL); |
Yasunori Goto | 6811378 | 2006-06-23 02:03:11 -0700 | [diff] [blame] | 3369 | /* cpuset refresh routine should be here */ |
| 3370 | } |
Andrew Morton | bd1e22b | 2006-06-23 02:03:47 -0700 | [diff] [blame] | 3371 | vm_total_pages = nr_free_pagecache_pages(); |
Mel Gorman | 9ef9acb | 2007-10-16 01:25:54 -0700 | [diff] [blame] | 3372 | /* |
| 3373 | * Disable grouping by mobility if the number of pages in the |
| 3374 | * system is too low to allow the mechanism to work. It would be |
| 3375 | * more accurate, but expensive to check per-zone. This check is |
| 3376 | * made on memory-hotadd so a system can start with mobility |
| 3377 | * disabled and enable it later |
| 3378 | */ |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 3379 | if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES)) |
Mel Gorman | 9ef9acb | 2007-10-16 01:25:54 -0700 | [diff] [blame] | 3380 | page_group_by_mobility_disabled = 1; |
| 3381 | else |
| 3382 | page_group_by_mobility_disabled = 0; |
| 3383 | |
| 3384 | printk("Built %i zonelists in %s order, mobility grouping %s. " |
| 3385 | "Total pages: %ld\n", |
Christoph Lameter | 62bc62a | 2009-06-16 15:32:15 -0700 | [diff] [blame] | 3386 | nr_online_nodes, |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3387 | zonelist_order_name[current_zonelist_order], |
Mel Gorman | 9ef9acb | 2007-10-16 01:25:54 -0700 | [diff] [blame] | 3388 | page_group_by_mobility_disabled ? "off" : "on", |
KAMEZAWA Hiroyuki | f0c0b2b | 2007-07-15 23:38:01 -0700 | [diff] [blame] | 3389 | vm_total_pages); |
| 3390 | #ifdef CONFIG_NUMA |
| 3391 | printk("Policy zone: %s\n", zone_names[policy_zone]); |
| 3392 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3393 | } |
| 3394 | |
| 3395 | /* |
| 3396 | * Helper functions to size the waitqueue hash table. |
| 3397 | * Essentially these want to choose hash table sizes sufficiently |
| 3398 | * large so that collisions trying to wait on pages are rare. |
| 3399 | * But in fact, the number of active page waitqueues on typical |
| 3400 | * systems is ridiculously low, less than 200. So this is even |
| 3401 | * conservative, even though it seems large. |
| 3402 | * |
| 3403 | * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to |
| 3404 | * waitqueues, i.e. the size of the waitq table given the number of pages. |
| 3405 | */ |
| 3406 | #define PAGES_PER_WAITQUEUE 256 |
| 3407 | |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3408 | #ifndef CONFIG_MEMORY_HOTPLUG |
Yasunori Goto | 02b694d | 2006-06-23 02:03:08 -0700 | [diff] [blame] | 3409 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3410 | { |
| 3411 | unsigned long size = 1; |
| 3412 | |
| 3413 | pages /= PAGES_PER_WAITQUEUE; |
| 3414 | |
| 3415 | while (size < pages) |
| 3416 | size <<= 1; |
| 3417 | |
| 3418 | /* |
| 3419 | * Once we have dozens or even hundreds of threads sleeping |
| 3420 | * on IO we've got bigger problems than wait queue collision. |
| 3421 | * Limit the size of the wait table to a reasonable size. |
| 3422 | */ |
| 3423 | size = min(size, 4096UL); |
| 3424 | |
| 3425 | return max(size, 4UL); |
| 3426 | } |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3427 | #else |
| 3428 | /* |
| 3429 | * A zone's size might be changed by hot-add, so it is not possible to determine |
| 3430 | * a suitable size for its wait_table. So we use the maximum size now. |
| 3431 | * |
| 3432 | * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie: |
| 3433 | * |
| 3434 | * i386 (preemption config) : 4096 x 16 = 64Kbyte. |
| 3435 | * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte. |
| 3436 | * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte. |
| 3437 | * |
| 3438 | * The maximum entries are prepared when a zone's memory is (512K + 256) pages |
| 3439 | * or more by the traditional way. (See above). It equals: |
| 3440 | * |
| 3441 | * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte. |
| 3442 | * ia64(16K page size) : = ( 8G + 4M)byte. |
| 3443 | * powerpc (64K page size) : = (32G +16M)byte. |
| 3444 | */ |
| 3445 | static inline unsigned long wait_table_hash_nr_entries(unsigned long pages) |
| 3446 | { |
| 3447 | return 4096UL; |
| 3448 | } |
| 3449 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3450 | |
| 3451 | /* |
| 3452 | * This is an integer logarithm so that shifts can be used later |
| 3453 | * to extract the more random high bits from the multiplicative |
| 3454 | * hash function before the remainder is taken. |
| 3455 | */ |
| 3456 | static inline unsigned long wait_table_bits(unsigned long size) |
| 3457 | { |
| 3458 | return ffz(~size); |
| 3459 | } |
| 3460 | |
| 3461 | #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1)) |
| 3462 | |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3463 | /* |
Arve Hjønnevåg | 6d3163c | 2011-05-24 17:12:24 -0700 | [diff] [blame] | 3464 | * Check if a pageblock contains reserved pages |
| 3465 | */ |
| 3466 | static int pageblock_is_reserved(unsigned long start_pfn, unsigned long end_pfn) |
| 3467 | { |
| 3468 | unsigned long pfn; |
| 3469 | |
| 3470 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
| 3471 | if (!pfn_valid_within(pfn) || PageReserved(pfn_to_page(pfn))) |
| 3472 | return 1; |
| 3473 | } |
| 3474 | return 0; |
| 3475 | } |
| 3476 | |
| 3477 | /* |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 3478 | * Mark a number of pageblocks as MIGRATE_RESERVE. The number |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 3479 | * of blocks reserved is based on min_wmark_pages(zone). The memory within |
| 3480 | * the reserve will tend to store contiguous free pages. Setting min_free_kbytes |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3481 | * higher will lead to a bigger reserve which will get freed as contiguous |
| 3482 | * blocks as reclaim kicks in |
| 3483 | */ |
| 3484 | static void setup_zone_migrate_reserve(struct zone *zone) |
| 3485 | { |
Arve Hjønnevåg | 6d3163c | 2011-05-24 17:12:24 -0700 | [diff] [blame] | 3486 | unsigned long start_pfn, pfn, end_pfn, block_end_pfn; |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3487 | struct page *page; |
Mel Gorman | 78986a6 | 2009-09-21 17:03:02 -0700 | [diff] [blame] | 3488 | unsigned long block_migratetype; |
| 3489 | int reserve; |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3490 | |
Michal Hocko | d0215638 | 2011-12-08 14:34:27 -0800 | [diff] [blame] | 3491 | /* |
| 3492 | * Get the start pfn, end pfn and the number of blocks to reserve |
| 3493 | * We have to be careful to be aligned to pageblock_nr_pages to |
| 3494 | * make sure that we always check pfn_valid for the first page in |
| 3495 | * the block. |
| 3496 | */ |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3497 | start_pfn = zone->zone_start_pfn; |
| 3498 | end_pfn = start_pfn + zone->spanned_pages; |
Michal Hocko | d0215638 | 2011-12-08 14:34:27 -0800 | [diff] [blame] | 3499 | start_pfn = roundup(start_pfn, pageblock_nr_pages); |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 3500 | reserve = roundup(min_wmark_pages(zone), pageblock_nr_pages) >> |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 3501 | pageblock_order; |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3502 | |
Mel Gorman | 78986a6 | 2009-09-21 17:03:02 -0700 | [diff] [blame] | 3503 | /* |
| 3504 | * Reserve blocks are generally in place to help high-order atomic |
| 3505 | * allocations that are short-lived. A min_free_kbytes value that |
| 3506 | * would result in more than 2 reserve blocks for atomic allocations |
| 3507 | * is assumed to be in place to help anti-fragmentation for the |
| 3508 | * future allocation of hugepages at runtime. |
| 3509 | */ |
| 3510 | reserve = min(2, reserve); |
| 3511 | |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 3512 | for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) { |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3513 | if (!pfn_valid(pfn)) |
| 3514 | continue; |
| 3515 | page = pfn_to_page(pfn); |
| 3516 | |
Adam Litke | 344c790 | 2008-09-02 14:35:38 -0700 | [diff] [blame] | 3517 | /* Watch out for overlapping nodes */ |
| 3518 | if (page_to_nid(page) != zone_to_nid(zone)) |
| 3519 | continue; |
| 3520 | |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3521 | block_migratetype = get_pageblock_migratetype(page); |
| 3522 | |
Mel Gorman | 938929f | 2012-01-10 15:07:14 -0800 | [diff] [blame] | 3523 | /* Only test what is necessary when the reserves are not met */ |
| 3524 | if (reserve > 0) { |
| 3525 | /* |
| 3526 | * Blocks with reserved pages will never free, skip |
| 3527 | * them. |
| 3528 | */ |
| 3529 | block_end_pfn = min(pfn + pageblock_nr_pages, end_pfn); |
| 3530 | if (pageblock_is_reserved(pfn, block_end_pfn)) |
| 3531 | continue; |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3532 | |
Mel Gorman | 938929f | 2012-01-10 15:07:14 -0800 | [diff] [blame] | 3533 | /* If this block is reserved, account for it */ |
| 3534 | if (block_migratetype == MIGRATE_RESERVE) { |
| 3535 | reserve--; |
| 3536 | continue; |
| 3537 | } |
| 3538 | |
| 3539 | /* Suitable for reserving if this block is movable */ |
| 3540 | if (block_migratetype == MIGRATE_MOVABLE) { |
| 3541 | set_pageblock_migratetype(page, |
| 3542 | MIGRATE_RESERVE); |
| 3543 | move_freepages_block(zone, page, |
| 3544 | MIGRATE_RESERVE); |
| 3545 | reserve--; |
| 3546 | continue; |
| 3547 | } |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3548 | } |
| 3549 | |
| 3550 | /* |
| 3551 | * If the reserve is met and this is a previous reserved block, |
| 3552 | * take it back |
| 3553 | */ |
| 3554 | if (block_migratetype == MIGRATE_RESERVE) { |
| 3555 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
| 3556 | move_freepages_block(zone, page, MIGRATE_MOVABLE); |
| 3557 | } |
| 3558 | } |
| 3559 | } |
Mel Gorman | ac0e5b7 | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3560 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3561 | /* |
| 3562 | * Initially all pages are reserved - free ones are freed |
| 3563 | * up by free_all_bootmem() once the early boot process is |
| 3564 | * done. Non-atomic initialization, single-pass. |
| 3565 | */ |
Matt Tolentino | c09b424 | 2006-01-17 07:03:44 +0100 | [diff] [blame] | 3566 | void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone, |
Dave Hansen | a2f3aa02 | 2007-01-10 23:15:30 -0800 | [diff] [blame] | 3567 | unsigned long start_pfn, enum memmap_context context) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3568 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3569 | struct page *page; |
Andy Whitcroft | 29751f6 | 2005-06-23 00:08:00 -0700 | [diff] [blame] | 3570 | unsigned long end_pfn = start_pfn + size; |
| 3571 | unsigned long pfn; |
KAMEZAWA Hiroyuki | 86051ca | 2008-04-29 00:58:21 -0700 | [diff] [blame] | 3572 | struct zone *z; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3573 | |
Hugh Dickins | 22b31ee | 2009-01-06 14:40:09 -0800 | [diff] [blame] | 3574 | if (highest_memmap_pfn < end_pfn - 1) |
| 3575 | highest_memmap_pfn = end_pfn - 1; |
| 3576 | |
KAMEZAWA Hiroyuki | 86051ca | 2008-04-29 00:58:21 -0700 | [diff] [blame] | 3577 | z = &NODE_DATA(nid)->node_zones[zone]; |
Greg Ungerer | cbe8dd4 | 2006-01-12 01:05:24 -0800 | [diff] [blame] | 3578 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { |
Dave Hansen | a2f3aa02 | 2007-01-10 23:15:30 -0800 | [diff] [blame] | 3579 | /* |
| 3580 | * There can be holes in boot-time mem_map[]s |
| 3581 | * handed to this function. They do not |
| 3582 | * exist on hotplugged memory. |
| 3583 | */ |
| 3584 | if (context == MEMMAP_EARLY) { |
| 3585 | if (!early_pfn_valid(pfn)) |
| 3586 | continue; |
| 3587 | if (!early_pfn_in_nid(pfn, nid)) |
| 3588 | continue; |
| 3589 | } |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 3590 | page = pfn_to_page(pfn); |
| 3591 | set_page_links(page, zone, nid, pfn); |
Mel Gorman | 708614e | 2008-07-23 21:26:51 -0700 | [diff] [blame] | 3592 | mminit_verify_page_links(page, zone, nid, pfn); |
Nick Piggin | 7835e98 | 2006-03-22 00:08:40 -0800 | [diff] [blame] | 3593 | init_page_count(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3594 | reset_page_mapcount(page); |
| 3595 | SetPageReserved(page); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 3596 | /* |
| 3597 | * Mark the block movable so that blocks are reserved for |
| 3598 | * movable at startup. This will force kernel allocations |
| 3599 | * to reserve their blocks rather than leaking throughout |
| 3600 | * the address space during boot when many long-lived |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3601 | * kernel allocations are made. Later some blocks near |
| 3602 | * the start are marked MIGRATE_RESERVE by |
| 3603 | * setup_zone_migrate_reserve() |
KAMEZAWA Hiroyuki | 86051ca | 2008-04-29 00:58:21 -0700 | [diff] [blame] | 3604 | * |
| 3605 | * bitmap is created for zone's valid pfn range. but memmap |
| 3606 | * can be created for invalid pages (for alignment) |
| 3607 | * check here not to call set_pageblock_migratetype() against |
| 3608 | * pfn out of zone. |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 3609 | */ |
KAMEZAWA Hiroyuki | 86051ca | 2008-04-29 00:58:21 -0700 | [diff] [blame] | 3610 | if ((z->zone_start_pfn <= pfn) |
| 3611 | && (pfn < z->zone_start_pfn + z->spanned_pages) |
| 3612 | && !(pfn & (pageblock_nr_pages - 1))) |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 3613 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 3614 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3615 | INIT_LIST_HEAD(&page->lru); |
| 3616 | #ifdef WANT_PAGE_VIRTUAL |
| 3617 | /* The shift won't overflow because ZONE_NORMAL is below 4G. */ |
| 3618 | if (!is_highmem_idx(zone)) |
Bob Picco | 3212c6b | 2005-06-27 14:36:28 -0700 | [diff] [blame] | 3619 | set_page_address(page, __va(pfn << PAGE_SHIFT)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3620 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3621 | } |
| 3622 | } |
| 3623 | |
Andi Kleen | 1e548de | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 3624 | static void __meminit zone_init_free_lists(struct zone *zone) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3625 | { |
Mel Gorman | b2a0ac8 | 2007-10-16 01:25:48 -0700 | [diff] [blame] | 3626 | int order, t; |
| 3627 | for_each_migratetype_order(order, t) { |
| 3628 | INIT_LIST_HEAD(&zone->free_area[order].free_list[t]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3629 | zone->free_area[order].nr_free = 0; |
| 3630 | } |
| 3631 | } |
| 3632 | |
| 3633 | #ifndef __HAVE_ARCH_MEMMAP_INIT |
| 3634 | #define memmap_init(size, nid, zone, start_pfn) \ |
Dave Hansen | a2f3aa02 | 2007-01-10 23:15:30 -0800 | [diff] [blame] | 3635 | memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3636 | #endif |
| 3637 | |
Sam Ravnborg | 1d6f4e6 | 2008-01-17 15:21:12 -0800 | [diff] [blame] | 3638 | static int zone_batchsize(struct zone *zone) |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3639 | { |
David Howells | 3a6be87 | 2009-05-06 16:03:03 -0700 | [diff] [blame] | 3640 | #ifdef CONFIG_MMU |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3641 | int batch; |
| 3642 | |
| 3643 | /* |
| 3644 | * The per-cpu-pages pools are set to around 1000th of the |
Seth, Rohit | ba56e91 | 2005-10-29 18:15:47 -0700 | [diff] [blame] | 3645 | * size of the zone. But no more than 1/2 of a meg. |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3646 | * |
| 3647 | * OK, so we don't know how big the cache is. So guess. |
| 3648 | */ |
| 3649 | batch = zone->present_pages / 1024; |
Seth, Rohit | ba56e91 | 2005-10-29 18:15:47 -0700 | [diff] [blame] | 3650 | if (batch * PAGE_SIZE > 512 * 1024) |
| 3651 | batch = (512 * 1024) / PAGE_SIZE; |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3652 | batch /= 4; /* We effectively *= 4 below */ |
| 3653 | if (batch < 1) |
| 3654 | batch = 1; |
| 3655 | |
| 3656 | /* |
Nick Piggin | 0ceaacc | 2005-12-04 13:55:25 +1100 | [diff] [blame] | 3657 | * Clamp the batch to a 2^n - 1 value. Having a power |
| 3658 | * of 2 value was found to be more likely to have |
| 3659 | * suboptimal cache aliasing properties in some cases. |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3660 | * |
Nick Piggin | 0ceaacc | 2005-12-04 13:55:25 +1100 | [diff] [blame] | 3661 | * For example if 2 tasks are alternately allocating |
| 3662 | * batches of pages, one task can end up with a lot |
| 3663 | * of pages of one half of the possible page colors |
| 3664 | * and the other with pages of the other colors. |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3665 | */ |
David Howells | 9155203 | 2009-05-06 16:03:02 -0700 | [diff] [blame] | 3666 | batch = rounddown_pow_of_two(batch + batch/2) - 1; |
Seth, Rohit | ba56e91 | 2005-10-29 18:15:47 -0700 | [diff] [blame] | 3667 | |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3668 | return batch; |
David Howells | 3a6be87 | 2009-05-06 16:03:03 -0700 | [diff] [blame] | 3669 | |
| 3670 | #else |
| 3671 | /* The deferral and batching of frees should be suppressed under NOMMU |
| 3672 | * conditions. |
| 3673 | * |
| 3674 | * The problem is that NOMMU needs to be able to allocate large chunks |
| 3675 | * of contiguous memory as there's no hardware page translation to |
| 3676 | * assemble apparent contiguous memory from discontiguous pages. |
| 3677 | * |
| 3678 | * Queueing large contiguous runs of pages for batching, however, |
| 3679 | * causes the pages to actually be freed in smaller chunks. As there |
| 3680 | * can be a significant delay between the individual batches being |
| 3681 | * recycled, this leads to the once large chunks of space being |
| 3682 | * fragmented and becoming unavailable for high-order allocations. |
| 3683 | */ |
| 3684 | return 0; |
| 3685 | #endif |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3686 | } |
| 3687 | |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 3688 | static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch) |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 3689 | { |
| 3690 | struct per_cpu_pages *pcp; |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 3691 | int migratetype; |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 3692 | |
Magnus Damm | 1c6fe94 | 2005-10-26 01:58:59 -0700 | [diff] [blame] | 3693 | memset(p, 0, sizeof(*p)); |
| 3694 | |
Christoph Lameter | 3dfa572 | 2008-02-04 22:29:19 -0800 | [diff] [blame] | 3695 | pcp = &p->pcp; |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 3696 | pcp->count = 0; |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 3697 | pcp->high = 6 * batch; |
| 3698 | pcp->batch = max(1UL, 1 * batch); |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 3699 | for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++) |
| 3700 | INIT_LIST_HEAD(&pcp->lists[migratetype]); |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 3701 | } |
| 3702 | |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 3703 | /* |
| 3704 | * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist |
| 3705 | * to the value high for the pageset p. |
| 3706 | */ |
| 3707 | |
| 3708 | static void setup_pagelist_highmark(struct per_cpu_pageset *p, |
| 3709 | unsigned long high) |
| 3710 | { |
| 3711 | struct per_cpu_pages *pcp; |
| 3712 | |
Christoph Lameter | 3dfa572 | 2008-02-04 22:29:19 -0800 | [diff] [blame] | 3713 | pcp = &p->pcp; |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 3714 | pcp->high = high; |
| 3715 | pcp->batch = max(1UL, high/4); |
| 3716 | if ((high/4) > (PAGE_SHIFT * 8)) |
| 3717 | pcp->batch = PAGE_SHIFT * 8; |
| 3718 | } |
| 3719 | |
Nikanth Karthikesan | 58c2ee4 | 2011-03-15 10:59:02 +0530 | [diff] [blame] | 3720 | static void setup_zone_pageset(struct zone *zone) |
Wu Fengguang | 319774e | 2010-05-24 14:32:49 -0700 | [diff] [blame] | 3721 | { |
| 3722 | int cpu; |
| 3723 | |
| 3724 | zone->pageset = alloc_percpu(struct per_cpu_pageset); |
| 3725 | |
| 3726 | for_each_possible_cpu(cpu) { |
| 3727 | struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu); |
| 3728 | |
| 3729 | setup_pageset(pcp, zone_batchsize(zone)); |
| 3730 | |
| 3731 | if (percpu_pagelist_fraction) |
| 3732 | setup_pagelist_highmark(pcp, |
| 3733 | (zone->present_pages / |
| 3734 | percpu_pagelist_fraction)); |
| 3735 | } |
| 3736 | } |
| 3737 | |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3738 | /* |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 3739 | * Allocate per cpu pagesets and initialize them. |
| 3740 | * Before this call only boot pagesets were available. |
Christoph Lameter | 2caaad4 | 2005-06-21 17:15:00 -0700 | [diff] [blame] | 3741 | */ |
Al Viro | 78d9955 | 2005-12-15 09:18:25 +0000 | [diff] [blame] | 3742 | void __init setup_per_cpu_pageset(void) |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3743 | { |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 3744 | struct zone *zone; |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3745 | |
Wu Fengguang | 319774e | 2010-05-24 14:32:49 -0700 | [diff] [blame] | 3746 | for_each_populated_zone(zone) |
| 3747 | setup_zone_pageset(zone); |
Christoph Lameter | e7c8d5c | 2005-06-21 17:14:47 -0700 | [diff] [blame] | 3748 | } |
| 3749 | |
Sam Ravnborg | 577a32f | 2007-05-17 23:29:25 +0200 | [diff] [blame] | 3750 | static noinline __init_refok |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3751 | int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages) |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3752 | { |
| 3753 | int i; |
| 3754 | struct pglist_data *pgdat = zone->zone_pgdat; |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3755 | size_t alloc_size; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3756 | |
| 3757 | /* |
| 3758 | * The per-page waitqueue mechanism uses hashed waitqueues |
| 3759 | * per zone. |
| 3760 | */ |
Yasunori Goto | 02b694d | 2006-06-23 02:03:08 -0700 | [diff] [blame] | 3761 | zone->wait_table_hash_nr_entries = |
| 3762 | wait_table_hash_nr_entries(zone_size_pages); |
| 3763 | zone->wait_table_bits = |
| 3764 | wait_table_bits(zone->wait_table_hash_nr_entries); |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3765 | alloc_size = zone->wait_table_hash_nr_entries |
| 3766 | * sizeof(wait_queue_head_t); |
| 3767 | |
Heiko Carstens | cd94b9d | 2008-05-23 13:04:52 -0700 | [diff] [blame] | 3768 | if (!slab_is_available()) { |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3769 | zone->wait_table = (wait_queue_head_t *) |
Yinghai Lu | 8f389a99 | 2011-05-11 15:13:32 -0700 | [diff] [blame] | 3770 | alloc_bootmem_node_nopanic(pgdat, alloc_size); |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3771 | } else { |
| 3772 | /* |
| 3773 | * This case means that a zone whose size was 0 gets new memory |
| 3774 | * via memory hot-add. |
| 3775 | * But it may be the case that a new node was hot-added. In |
| 3776 | * this case vmalloc() will not be able to use this new node's |
| 3777 | * memory - this wait_table must be initialized to use this new |
| 3778 | * node itself as well. |
| 3779 | * To use this new node's memory, further consideration will be |
| 3780 | * necessary. |
| 3781 | */ |
Jesper Juhl | 8691f3a | 2007-10-16 01:24:49 -0700 | [diff] [blame] | 3782 | zone->wait_table = vmalloc(alloc_size); |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3783 | } |
| 3784 | if (!zone->wait_table) |
| 3785 | return -ENOMEM; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3786 | |
Yasunori Goto | 02b694d | 2006-06-23 02:03:08 -0700 | [diff] [blame] | 3787 | for(i = 0; i < zone->wait_table_hash_nr_entries; ++i) |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3788 | init_waitqueue_head(zone->wait_table + i); |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3789 | |
| 3790 | return 0; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3791 | } |
| 3792 | |
Shaohua Li | 112067f | 2009-09-21 17:01:16 -0700 | [diff] [blame] | 3793 | static int __zone_pcp_update(void *data) |
| 3794 | { |
| 3795 | struct zone *zone = data; |
| 3796 | int cpu; |
| 3797 | unsigned long batch = zone_batchsize(zone), flags; |
| 3798 | |
Thomas Gleixner | 2d30a1f | 2010-03-10 15:20:40 -0800 | [diff] [blame] | 3799 | for_each_possible_cpu(cpu) { |
Shaohua Li | 112067f | 2009-09-21 17:01:16 -0700 | [diff] [blame] | 3800 | struct per_cpu_pageset *pset; |
| 3801 | struct per_cpu_pages *pcp; |
| 3802 | |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 3803 | pset = per_cpu_ptr(zone->pageset, cpu); |
Shaohua Li | 112067f | 2009-09-21 17:01:16 -0700 | [diff] [blame] | 3804 | pcp = &pset->pcp; |
| 3805 | |
| 3806 | local_irq_save(flags); |
Mel Gorman | 5f8dcc2 | 2009-09-21 17:03:19 -0700 | [diff] [blame] | 3807 | free_pcppages_bulk(zone, pcp->count, pcp); |
Shaohua Li | 112067f | 2009-09-21 17:01:16 -0700 | [diff] [blame] | 3808 | setup_pageset(pset, batch); |
| 3809 | local_irq_restore(flags); |
| 3810 | } |
| 3811 | return 0; |
| 3812 | } |
| 3813 | |
| 3814 | void zone_pcp_update(struct zone *zone) |
| 3815 | { |
| 3816 | stop_machine(__zone_pcp_update, zone, NULL); |
| 3817 | } |
| 3818 | |
Matt Tolentino | c09b424 | 2006-01-17 07:03:44 +0100 | [diff] [blame] | 3819 | static __meminit void zone_pcp_init(struct zone *zone) |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3820 | { |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 3821 | /* |
| 3822 | * per cpu subsystem is not up at this point. The following code |
| 3823 | * relies on the ability of the linker to provide the |
| 3824 | * offset of a (static) per cpu variable into the per cpu area. |
| 3825 | */ |
| 3826 | zone->pageset = &boot_pageset; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3827 | |
Anton Blanchard | f5335c0 | 2006-03-25 03:06:49 -0800 | [diff] [blame] | 3828 | if (zone->present_pages) |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 3829 | printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n", |
| 3830 | zone->name, zone->present_pages, |
| 3831 | zone_batchsize(zone)); |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3832 | } |
| 3833 | |
Yasunori Goto | 718127c | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3834 | __meminit int init_currently_empty_zone(struct zone *zone, |
| 3835 | unsigned long zone_start_pfn, |
Dave Hansen | a2f3aa02 | 2007-01-10 23:15:30 -0800 | [diff] [blame] | 3836 | unsigned long size, |
| 3837 | enum memmap_context context) |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3838 | { |
| 3839 | struct pglist_data *pgdat = zone->zone_pgdat; |
Yasunori Goto | cca448f | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3840 | int ret; |
| 3841 | ret = zone_wait_table_init(zone, size); |
| 3842 | if (ret) |
| 3843 | return ret; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3844 | pgdat->nr_zones = zone_idx(zone) + 1; |
| 3845 | |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3846 | zone->zone_start_pfn = zone_start_pfn; |
| 3847 | |
Mel Gorman | 708614e | 2008-07-23 21:26:51 -0700 | [diff] [blame] | 3848 | mminit_dprintk(MMINIT_TRACE, "memmap_init", |
| 3849 | "Initialising map node %d zone %lu pfns %lu -> %lu\n", |
| 3850 | pgdat->node_id, |
| 3851 | (unsigned long)zone_idx(zone), |
| 3852 | zone_start_pfn, (zone_start_pfn + size)); |
| 3853 | |
Andi Kleen | 1e548de | 2008-02-04 22:29:26 -0800 | [diff] [blame] | 3854 | zone_init_free_lists(zone); |
Yasunori Goto | 718127c | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 3855 | |
| 3856 | return 0; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 3857 | } |
| 3858 | |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 3859 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3860 | #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID |
| 3861 | /* |
| 3862 | * Required by SPARSEMEM. Given a PFN, return what node the PFN is on. |
| 3863 | * Architectures may implement their own version but if add_active_range() |
| 3864 | * was used and there are no special requirements, this is a convenient |
| 3865 | * alternative |
| 3866 | */ |
KAMEZAWA Hiroyuki | f2dbcfa | 2009-02-18 14:48:32 -0800 | [diff] [blame] | 3867 | int __meminit __early_pfn_to_nid(unsigned long pfn) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3868 | { |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 3869 | unsigned long start_pfn, end_pfn; |
| 3870 | int i, nid; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3871 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 3872 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3873 | if (start_pfn <= pfn && pfn < end_pfn) |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 3874 | return nid; |
KAMEZAWA Hiroyuki | cc2559b | 2009-02-18 14:48:33 -0800 | [diff] [blame] | 3875 | /* This is a memory hole */ |
| 3876 | return -1; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3877 | } |
| 3878 | #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */ |
| 3879 | |
KAMEZAWA Hiroyuki | f2dbcfa | 2009-02-18 14:48:32 -0800 | [diff] [blame] | 3880 | int __meminit early_pfn_to_nid(unsigned long pfn) |
| 3881 | { |
KAMEZAWA Hiroyuki | cc2559b | 2009-02-18 14:48:33 -0800 | [diff] [blame] | 3882 | int nid; |
| 3883 | |
| 3884 | nid = __early_pfn_to_nid(pfn); |
| 3885 | if (nid >= 0) |
| 3886 | return nid; |
| 3887 | /* just returns 0 */ |
| 3888 | return 0; |
KAMEZAWA Hiroyuki | f2dbcfa | 2009-02-18 14:48:32 -0800 | [diff] [blame] | 3889 | } |
| 3890 | |
KAMEZAWA Hiroyuki | cc2559b | 2009-02-18 14:48:33 -0800 | [diff] [blame] | 3891 | #ifdef CONFIG_NODES_SPAN_OTHER_NODES |
| 3892 | bool __meminit early_pfn_in_nid(unsigned long pfn, int node) |
| 3893 | { |
| 3894 | int nid; |
| 3895 | |
| 3896 | nid = __early_pfn_to_nid(pfn); |
| 3897 | if (nid >= 0 && nid != node) |
| 3898 | return false; |
| 3899 | return true; |
| 3900 | } |
| 3901 | #endif |
KAMEZAWA Hiroyuki | f2dbcfa | 2009-02-18 14:48:32 -0800 | [diff] [blame] | 3902 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3903 | /** |
| 3904 | * free_bootmem_with_active_regions - Call free_bootmem_node for each active range |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 3905 | * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed. |
| 3906 | * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3907 | * |
| 3908 | * If an architecture guarantees that all ranges registered with |
| 3909 | * add_active_ranges() contain no holes and may be freed, this |
| 3910 | * this function may be used instead of calling free_bootmem() manually. |
| 3911 | */ |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 3912 | void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3913 | { |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 3914 | unsigned long start_pfn, end_pfn; |
| 3915 | int i, this_nid; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3916 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 3917 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) { |
| 3918 | start_pfn = min(start_pfn, max_low_pfn); |
| 3919 | end_pfn = min(end_pfn, max_low_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3920 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 3921 | if (start_pfn < end_pfn) |
| 3922 | free_bootmem_node(NODE_DATA(this_nid), |
| 3923 | PFN_PHYS(start_pfn), |
| 3924 | (end_pfn - start_pfn) << PAGE_SHIFT); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3925 | } |
| 3926 | } |
| 3927 | |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 3928 | int __init add_from_early_node_map(struct range *range, int az, |
| 3929 | int nr_range, int nid) |
| 3930 | { |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 3931 | unsigned long start_pfn, end_pfn; |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 3932 | int i; |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 3933 | |
| 3934 | /* need to go over early_node_map to find out good range for node */ |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 3935 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) |
| 3936 | nr_range = add_range(range, az, nr_range, start_pfn, end_pfn); |
Yinghai Lu | 0867721 | 2010-02-10 01:20:20 -0800 | [diff] [blame] | 3937 | return nr_range; |
| 3938 | } |
| 3939 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3940 | /** |
| 3941 | * sparse_memory_present_with_active_regions - Call memory_present for each active range |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 3942 | * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used. |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3943 | * |
| 3944 | * If an architecture guarantees that all ranges registered with |
| 3945 | * add_active_ranges() contain no holes and may be freed, this |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 3946 | * function may be used instead of calling memory_present() manually. |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3947 | */ |
| 3948 | void __init sparse_memory_present_with_active_regions(int nid) |
| 3949 | { |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 3950 | unsigned long start_pfn, end_pfn; |
| 3951 | int i, this_nid; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3952 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 3953 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) |
| 3954 | memory_present(this_nid, start_pfn, end_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3955 | } |
| 3956 | |
| 3957 | /** |
| 3958 | * get_pfn_range_for_nid - Return the start and end page frames for a node |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 3959 | * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned. |
| 3960 | * @start_pfn: Passed by reference. On return, it will have the node start_pfn. |
| 3961 | * @end_pfn: Passed by reference. On return, it will have the node end_pfn. |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3962 | * |
| 3963 | * It returns the start and end page frame of a node based on information |
| 3964 | * provided by an arch calling add_active_range(). If called for a node |
| 3965 | * with no available memory, a warning is printed and the start and end |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 3966 | * PFNs will be 0. |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3967 | */ |
Yasunori Goto | a3142c8 | 2007-05-08 00:23:07 -0700 | [diff] [blame] | 3968 | void __meminit get_pfn_range_for_nid(unsigned int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3969 | unsigned long *start_pfn, unsigned long *end_pfn) |
| 3970 | { |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 3971 | unsigned long this_start_pfn, this_end_pfn; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3972 | int i; |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 3973 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3974 | *start_pfn = -1UL; |
| 3975 | *end_pfn = 0; |
| 3976 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 3977 | for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) { |
| 3978 | *start_pfn = min(*start_pfn, this_start_pfn); |
| 3979 | *end_pfn = max(*end_pfn, this_end_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3980 | } |
| 3981 | |
Christoph Lameter | 633c066 | 2007-10-16 01:25:37 -0700 | [diff] [blame] | 3982 | if (*start_pfn == -1UL) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3983 | *start_pfn = 0; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 3984 | } |
| 3985 | |
| 3986 | /* |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 3987 | * This finds a zone that can be used for ZONE_MOVABLE pages. The |
| 3988 | * assumption is made that zones within a node are ordered in monotonic |
| 3989 | * increasing memory addresses so that the "highest" populated zone is used |
| 3990 | */ |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 3991 | static void __init find_usable_zone_for_movable(void) |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 3992 | { |
| 3993 | int zone_index; |
| 3994 | for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) { |
| 3995 | if (zone_index == ZONE_MOVABLE) |
| 3996 | continue; |
| 3997 | |
| 3998 | if (arch_zone_highest_possible_pfn[zone_index] > |
| 3999 | arch_zone_lowest_possible_pfn[zone_index]) |
| 4000 | break; |
| 4001 | } |
| 4002 | |
| 4003 | VM_BUG_ON(zone_index == -1); |
| 4004 | movable_zone = zone_index; |
| 4005 | } |
| 4006 | |
| 4007 | /* |
| 4008 | * The zone ranges provided by the architecture do not include ZONE_MOVABLE |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 4009 | * because it is sized independent of architecture. Unlike the other zones, |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4010 | * the starting point for ZONE_MOVABLE is not fixed. It may be different |
| 4011 | * in each node depending on the size of each node and how evenly kernelcore |
| 4012 | * is distributed. This helper function adjusts the zone ranges |
| 4013 | * provided by the architecture for a given node by using the end of the |
| 4014 | * highest usable zone for ZONE_MOVABLE. This preserves the assumption that |
| 4015 | * zones within a node are in order of monotonic increases memory addresses |
| 4016 | */ |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 4017 | static void __meminit adjust_zone_range_for_zone_movable(int nid, |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4018 | unsigned long zone_type, |
| 4019 | unsigned long node_start_pfn, |
| 4020 | unsigned long node_end_pfn, |
| 4021 | unsigned long *zone_start_pfn, |
| 4022 | unsigned long *zone_end_pfn) |
| 4023 | { |
| 4024 | /* Only adjust if ZONE_MOVABLE is on this node */ |
| 4025 | if (zone_movable_pfn[nid]) { |
| 4026 | /* Size ZONE_MOVABLE */ |
| 4027 | if (zone_type == ZONE_MOVABLE) { |
| 4028 | *zone_start_pfn = zone_movable_pfn[nid]; |
| 4029 | *zone_end_pfn = min(node_end_pfn, |
| 4030 | arch_zone_highest_possible_pfn[movable_zone]); |
| 4031 | |
| 4032 | /* Adjust for ZONE_MOVABLE starting within this range */ |
| 4033 | } else if (*zone_start_pfn < zone_movable_pfn[nid] && |
| 4034 | *zone_end_pfn > zone_movable_pfn[nid]) { |
| 4035 | *zone_end_pfn = zone_movable_pfn[nid]; |
| 4036 | |
| 4037 | /* Check if this whole range is within ZONE_MOVABLE */ |
| 4038 | } else if (*zone_start_pfn >= zone_movable_pfn[nid]) |
| 4039 | *zone_start_pfn = *zone_end_pfn; |
| 4040 | } |
| 4041 | } |
| 4042 | |
| 4043 | /* |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4044 | * Return the number of pages a zone spans in a node, including holes |
| 4045 | * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node() |
| 4046 | */ |
Paul Mundt | 6ea6e68 | 2007-07-15 23:38:20 -0700 | [diff] [blame] | 4047 | static unsigned long __meminit zone_spanned_pages_in_node(int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4048 | unsigned long zone_type, |
| 4049 | unsigned long *ignored) |
| 4050 | { |
| 4051 | unsigned long node_start_pfn, node_end_pfn; |
| 4052 | unsigned long zone_start_pfn, zone_end_pfn; |
| 4053 | |
| 4054 | /* Get the start and end of the node and zone */ |
| 4055 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); |
| 4056 | zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type]; |
| 4057 | zone_end_pfn = arch_zone_highest_possible_pfn[zone_type]; |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4058 | adjust_zone_range_for_zone_movable(nid, zone_type, |
| 4059 | node_start_pfn, node_end_pfn, |
| 4060 | &zone_start_pfn, &zone_end_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4061 | |
| 4062 | /* Check that this node has pages within the zone's required range */ |
| 4063 | if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn) |
| 4064 | return 0; |
| 4065 | |
| 4066 | /* Move the zone boundaries inside the node if necessary */ |
| 4067 | zone_end_pfn = min(zone_end_pfn, node_end_pfn); |
| 4068 | zone_start_pfn = max(zone_start_pfn, node_start_pfn); |
| 4069 | |
| 4070 | /* Return the spanned pages */ |
| 4071 | return zone_end_pfn - zone_start_pfn; |
| 4072 | } |
| 4073 | |
| 4074 | /* |
| 4075 | * Return the number of holes in a range on a node. If nid is MAX_NUMNODES, |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 4076 | * then all holes in the requested range will be accounted for. |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4077 | */ |
Yinghai Lu | 3299625 | 2009-12-15 17:59:02 -0800 | [diff] [blame] | 4078 | unsigned long __meminit __absent_pages_in_range(int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4079 | unsigned long range_start_pfn, |
| 4080 | unsigned long range_end_pfn) |
| 4081 | { |
Tejun Heo | 96e907d | 2011-07-12 10:46:29 +0200 | [diff] [blame] | 4082 | unsigned long nr_absent = range_end_pfn - range_start_pfn; |
| 4083 | unsigned long start_pfn, end_pfn; |
| 4084 | int i; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4085 | |
Tejun Heo | 96e907d | 2011-07-12 10:46:29 +0200 | [diff] [blame] | 4086 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
| 4087 | start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn); |
| 4088 | end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn); |
| 4089 | nr_absent -= end_pfn - start_pfn; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4090 | } |
Tejun Heo | 96e907d | 2011-07-12 10:46:29 +0200 | [diff] [blame] | 4091 | return nr_absent; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4092 | } |
| 4093 | |
| 4094 | /** |
| 4095 | * absent_pages_in_range - Return number of page frames in holes within a range |
| 4096 | * @start_pfn: The start PFN to start searching for holes |
| 4097 | * @end_pfn: The end PFN to stop searching for holes |
| 4098 | * |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 4099 | * It returns the number of pages frames in memory holes within a range. |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4100 | */ |
| 4101 | unsigned long __init absent_pages_in_range(unsigned long start_pfn, |
| 4102 | unsigned long end_pfn) |
| 4103 | { |
| 4104 | return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn); |
| 4105 | } |
| 4106 | |
| 4107 | /* Return the number of page frames in holes in a zone on a node */ |
Paul Mundt | 6ea6e68 | 2007-07-15 23:38:20 -0700 | [diff] [blame] | 4108 | static unsigned long __meminit zone_absent_pages_in_node(int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4109 | unsigned long zone_type, |
| 4110 | unsigned long *ignored) |
| 4111 | { |
Tejun Heo | 96e907d | 2011-07-12 10:46:29 +0200 | [diff] [blame] | 4112 | unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type]; |
| 4113 | unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type]; |
Mel Gorman | 9c7cd68 | 2006-09-27 01:49:58 -0700 | [diff] [blame] | 4114 | unsigned long node_start_pfn, node_end_pfn; |
| 4115 | unsigned long zone_start_pfn, zone_end_pfn; |
| 4116 | |
| 4117 | get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn); |
Tejun Heo | 96e907d | 2011-07-12 10:46:29 +0200 | [diff] [blame] | 4118 | zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high); |
| 4119 | zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high); |
Mel Gorman | 9c7cd68 | 2006-09-27 01:49:58 -0700 | [diff] [blame] | 4120 | |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4121 | adjust_zone_range_for_zone_movable(nid, zone_type, |
| 4122 | node_start_pfn, node_end_pfn, |
| 4123 | &zone_start_pfn, &zone_end_pfn); |
Mel Gorman | 9c7cd68 | 2006-09-27 01:49:58 -0700 | [diff] [blame] | 4124 | return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4125 | } |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 4126 | |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 4127 | #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
Paul Mundt | 6ea6e68 | 2007-07-15 23:38:20 -0700 | [diff] [blame] | 4128 | static inline unsigned long __meminit zone_spanned_pages_in_node(int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4129 | unsigned long zone_type, |
| 4130 | unsigned long *zones_size) |
| 4131 | { |
| 4132 | return zones_size[zone_type]; |
| 4133 | } |
| 4134 | |
Paul Mundt | 6ea6e68 | 2007-07-15 23:38:20 -0700 | [diff] [blame] | 4135 | static inline unsigned long __meminit zone_absent_pages_in_node(int nid, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4136 | unsigned long zone_type, |
| 4137 | unsigned long *zholes_size) |
| 4138 | { |
| 4139 | if (!zholes_size) |
| 4140 | return 0; |
| 4141 | |
| 4142 | return zholes_size[zone_type]; |
| 4143 | } |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 4144 | |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 4145 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4146 | |
Yasunori Goto | a3142c8 | 2007-05-08 00:23:07 -0700 | [diff] [blame] | 4147 | static void __meminit calculate_node_totalpages(struct pglist_data *pgdat, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4148 | unsigned long *zones_size, unsigned long *zholes_size) |
| 4149 | { |
| 4150 | unsigned long realtotalpages, totalpages = 0; |
| 4151 | enum zone_type i; |
| 4152 | |
| 4153 | for (i = 0; i < MAX_NR_ZONES; i++) |
| 4154 | totalpages += zone_spanned_pages_in_node(pgdat->node_id, i, |
| 4155 | zones_size); |
| 4156 | pgdat->node_spanned_pages = totalpages; |
| 4157 | |
| 4158 | realtotalpages = totalpages; |
| 4159 | for (i = 0; i < MAX_NR_ZONES; i++) |
| 4160 | realtotalpages -= |
| 4161 | zone_absent_pages_in_node(pgdat->node_id, i, |
| 4162 | zholes_size); |
| 4163 | pgdat->node_present_pages = realtotalpages; |
| 4164 | printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id, |
| 4165 | realtotalpages); |
| 4166 | } |
| 4167 | |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 4168 | #ifndef CONFIG_SPARSEMEM |
| 4169 | /* |
| 4170 | * Calculate the size of the zone->blockflags rounded to an unsigned long |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 4171 | * Start by making sure zonesize is a multiple of pageblock_order by rounding |
| 4172 | * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 4173 | * round what is now in bits to nearest long in bits, then return it in |
| 4174 | * bytes. |
| 4175 | */ |
| 4176 | static unsigned long __init usemap_size(unsigned long zonesize) |
| 4177 | { |
| 4178 | unsigned long usemapsize; |
| 4179 | |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 4180 | usemapsize = roundup(zonesize, pageblock_nr_pages); |
| 4181 | usemapsize = usemapsize >> pageblock_order; |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 4182 | usemapsize *= NR_PAGEBLOCK_BITS; |
| 4183 | usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long)); |
| 4184 | |
| 4185 | return usemapsize / 8; |
| 4186 | } |
| 4187 | |
| 4188 | static void __init setup_usemap(struct pglist_data *pgdat, |
| 4189 | struct zone *zone, unsigned long zonesize) |
| 4190 | { |
| 4191 | unsigned long usemapsize = usemap_size(zonesize); |
| 4192 | zone->pageblock_flags = NULL; |
Julia Lawall | 58a01a4 | 2009-01-06 14:39:28 -0800 | [diff] [blame] | 4193 | if (usemapsize) |
Yinghai Lu | 8f389a99 | 2011-05-11 15:13:32 -0700 | [diff] [blame] | 4194 | zone->pageblock_flags = alloc_bootmem_node_nopanic(pgdat, |
| 4195 | usemapsize); |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 4196 | } |
| 4197 | #else |
Jesper Juhl | fa9f90b | 2010-11-28 21:39:34 +0100 | [diff] [blame] | 4198 | static inline void setup_usemap(struct pglist_data *pgdat, |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 4199 | struct zone *zone, unsigned long zonesize) {} |
| 4200 | #endif /* CONFIG_SPARSEMEM */ |
| 4201 | |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 4202 | #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE |
Mel Gorman | ba72cb8 | 2007-11-28 16:21:13 -0800 | [diff] [blame] | 4203 | |
| 4204 | /* Return a sensible default order for the pageblock size. */ |
| 4205 | static inline int pageblock_default_order(void) |
| 4206 | { |
| 4207 | if (HPAGE_SHIFT > PAGE_SHIFT) |
| 4208 | return HUGETLB_PAGE_ORDER; |
| 4209 | |
| 4210 | return MAX_ORDER-1; |
| 4211 | } |
| 4212 | |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 4213 | /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */ |
| 4214 | static inline void __init set_pageblock_order(unsigned int order) |
| 4215 | { |
| 4216 | /* Check that pageblock_nr_pages has not already been setup */ |
| 4217 | if (pageblock_order) |
| 4218 | return; |
| 4219 | |
| 4220 | /* |
| 4221 | * Assume the largest contiguous order of interest is a huge page. |
| 4222 | * This value may be variable depending on boot parameters on IA64 |
| 4223 | */ |
| 4224 | pageblock_order = order; |
| 4225 | } |
| 4226 | #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ |
| 4227 | |
Mel Gorman | ba72cb8 | 2007-11-28 16:21:13 -0800 | [diff] [blame] | 4228 | /* |
| 4229 | * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order() |
| 4230 | * and pageblock_default_order() are unused as pageblock_order is set |
| 4231 | * at compile-time. See include/linux/pageblock-flags.h for the values of |
| 4232 | * pageblock_order based on the kernel config |
| 4233 | */ |
| 4234 | static inline int pageblock_default_order(unsigned int order) |
| 4235 | { |
| 4236 | return MAX_ORDER-1; |
| 4237 | } |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 4238 | #define set_pageblock_order(x) do {} while (0) |
| 4239 | |
| 4240 | #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */ |
| 4241 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4242 | /* |
| 4243 | * Set up the zone data structures: |
| 4244 | * - mark all pages reserved |
| 4245 | * - mark all memory queues empty |
| 4246 | * - clear the memory bitmaps |
| 4247 | */ |
Alexander van Heukelum | b5a0e01 | 2008-02-23 15:24:06 -0800 | [diff] [blame] | 4248 | static void __paginginit free_area_init_core(struct pglist_data *pgdat, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4249 | unsigned long *zones_size, unsigned long *zholes_size) |
| 4250 | { |
Christoph Lameter | 2f1b624 | 2006-09-25 23:31:13 -0700 | [diff] [blame] | 4251 | enum zone_type j; |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 4252 | int nid = pgdat->node_id; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4253 | unsigned long zone_start_pfn = pgdat->node_start_pfn; |
Yasunori Goto | 718127c | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 4254 | int ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4255 | |
Dave Hansen | 208d54e | 2005-10-29 18:16:52 -0700 | [diff] [blame] | 4256 | pgdat_resize_init(pgdat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4257 | pgdat->nr_zones = 0; |
| 4258 | init_waitqueue_head(&pgdat->kswapd_wait); |
| 4259 | pgdat->kswapd_max_order = 0; |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 4260 | pgdat_page_cgroup_init(pgdat); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4261 | |
| 4262 | for (j = 0; j < MAX_NR_ZONES; j++) { |
| 4263 | struct zone *zone = pgdat->node_zones + j; |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 4264 | unsigned long size, realsize, memmap_pages; |
Hugh Dickins | 4111304 | 2012-01-12 17:20:01 -0800 | [diff] [blame^] | 4265 | enum lru_list lru; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4266 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4267 | size = zone_spanned_pages_in_node(nid, j, zones_size); |
| 4268 | realsize = size - zone_absent_pages_in_node(nid, j, |
| 4269 | zholes_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4270 | |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 4271 | /* |
| 4272 | * Adjust realsize so that it accounts for how much memory |
| 4273 | * is used by this zone for memmap. This affects the watermark |
| 4274 | * and per-cpu initialisations |
| 4275 | */ |
Johannes Weiner | f723215 | 2008-05-23 13:04:21 -0700 | [diff] [blame] | 4276 | memmap_pages = |
| 4277 | PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT; |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 4278 | if (realsize >= memmap_pages) { |
| 4279 | realsize -= memmap_pages; |
Yinghai Lu | 5594c8c | 2009-01-06 14:39:14 -0800 | [diff] [blame] | 4280 | if (memmap_pages) |
| 4281 | printk(KERN_DEBUG |
| 4282 | " %s zone: %lu pages used for memmap\n", |
| 4283 | zone_names[j], memmap_pages); |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 4284 | } else |
| 4285 | printk(KERN_WARNING |
| 4286 | " %s zone: %lu pages exceeds realsize %lu\n", |
| 4287 | zone_names[j], memmap_pages, realsize); |
| 4288 | |
Christoph Lameter | 6267276 | 2007-02-10 01:43:07 -0800 | [diff] [blame] | 4289 | /* Account for reserved pages */ |
| 4290 | if (j == 0 && realsize > dma_reserve) { |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 4291 | realsize -= dma_reserve; |
Yinghai Lu | d903ef9 | 2008-10-18 20:27:06 -0700 | [diff] [blame] | 4292 | printk(KERN_DEBUG " %s zone: %lu pages reserved\n", |
Christoph Lameter | 6267276 | 2007-02-10 01:43:07 -0800 | [diff] [blame] | 4293 | zone_names[0], dma_reserve); |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 4294 | } |
| 4295 | |
Christoph Lameter | 98d2b0e | 2006-09-25 23:31:12 -0700 | [diff] [blame] | 4296 | if (!is_highmem_idx(j)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4297 | nr_kernel_pages += realsize; |
| 4298 | nr_all_pages += realsize; |
| 4299 | |
| 4300 | zone->spanned_pages = size; |
| 4301 | zone->present_pages = realsize; |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 4302 | #ifdef CONFIG_NUMA |
Christoph Lameter | d5f541e | 2006-09-27 01:50:08 -0700 | [diff] [blame] | 4303 | zone->node = nid; |
Christoph Lameter | 8417bba | 2006-09-25 23:31:51 -0700 | [diff] [blame] | 4304 | zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio) |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 4305 | / 100; |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 4306 | zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100; |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 4307 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4308 | zone->name = zone_names[j]; |
| 4309 | spin_lock_init(&zone->lock); |
| 4310 | spin_lock_init(&zone->lru_lock); |
Dave Hansen | bdc8cb9 | 2005-10-29 18:16:53 -0700 | [diff] [blame] | 4311 | zone_seqlock_init(zone); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4312 | zone->zone_pgdat = pgdat; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4313 | |
Dave Hansen | ed8ece2 | 2005-10-29 18:16:50 -0700 | [diff] [blame] | 4314 | zone_pcp_init(zone); |
Hugh Dickins | 4111304 | 2012-01-12 17:20:01 -0800 | [diff] [blame^] | 4315 | for_each_lru(lru) |
| 4316 | INIT_LIST_HEAD(&zone->lruvec.lists[lru]); |
KOSAKI Motohiro | 6e90157 | 2009-01-07 18:08:15 -0800 | [diff] [blame] | 4317 | zone->reclaim_stat.recent_rotated[0] = 0; |
| 4318 | zone->reclaim_stat.recent_rotated[1] = 0; |
| 4319 | zone->reclaim_stat.recent_scanned[0] = 0; |
| 4320 | zone->reclaim_stat.recent_scanned[1] = 0; |
Christoph Lameter | 2244b95 | 2006-06-30 01:55:33 -0700 | [diff] [blame] | 4321 | zap_zone_vm_stats(zone); |
David Rientjes | e815af9 | 2007-10-16 23:25:54 -0700 | [diff] [blame] | 4322 | zone->flags = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4323 | if (!size) |
| 4324 | continue; |
| 4325 | |
Mel Gorman | ba72cb8 | 2007-11-28 16:21:13 -0800 | [diff] [blame] | 4326 | set_pageblock_order(pageblock_default_order()); |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 4327 | setup_usemap(pgdat, zone, size); |
Dave Hansen | a2f3aa02 | 2007-01-10 23:15:30 -0800 | [diff] [blame] | 4328 | ret = init_currently_empty_zone(zone, zone_start_pfn, |
| 4329 | size, MEMMAP_EARLY); |
Yasunori Goto | 718127c | 2006-06-23 02:03:10 -0700 | [diff] [blame] | 4330 | BUG_ON(ret); |
Heiko Carstens | 76cdd58 | 2008-05-14 16:05:52 -0700 | [diff] [blame] | 4331 | memmap_init(size, nid, j, zone_start_pfn); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4332 | zone_start_pfn += size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4333 | } |
| 4334 | } |
| 4335 | |
Sam Ravnborg | 577a32f | 2007-05-17 23:29:25 +0200 | [diff] [blame] | 4336 | static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4337 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4338 | /* Skip empty nodes */ |
| 4339 | if (!pgdat->node_spanned_pages) |
| 4340 | return; |
| 4341 | |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 4342 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4343 | /* ia64 gets its own node_mem_map, before this, without bootmem */ |
| 4344 | if (!pgdat->node_mem_map) { |
Bob Picco | e984bb4 | 2006-05-20 15:00:31 -0700 | [diff] [blame] | 4345 | unsigned long size, start, end; |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 4346 | struct page *map; |
| 4347 | |
Bob Picco | e984bb4 | 2006-05-20 15:00:31 -0700 | [diff] [blame] | 4348 | /* |
| 4349 | * The zone's endpoints aren't required to be MAX_ORDER |
| 4350 | * aligned but the node_mem_map endpoints must be in order |
| 4351 | * for the buddy allocator to function correctly. |
| 4352 | */ |
| 4353 | start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1); |
| 4354 | end = pgdat->node_start_pfn + pgdat->node_spanned_pages; |
| 4355 | end = ALIGN(end, MAX_ORDER_NR_PAGES); |
| 4356 | size = (end - start) * sizeof(struct page); |
Dave Hansen | 6f167ec | 2005-06-23 00:07:39 -0700 | [diff] [blame] | 4357 | map = alloc_remap(pgdat->node_id, size); |
| 4358 | if (!map) |
Yinghai Lu | 8f389a99 | 2011-05-11 15:13:32 -0700 | [diff] [blame] | 4359 | map = alloc_bootmem_node_nopanic(pgdat, size); |
Bob Picco | e984bb4 | 2006-05-20 15:00:31 -0700 | [diff] [blame] | 4360 | pgdat->node_mem_map = map + (pgdat->node_start_pfn - start); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4361 | } |
Roman Zippel | 12d810c | 2007-05-31 00:40:54 -0700 | [diff] [blame] | 4362 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4363 | /* |
| 4364 | * With no DISCONTIG, the global mem_map is just set as node 0's |
| 4365 | */ |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4366 | if (pgdat == NODE_DATA(0)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4367 | mem_map = NODE_DATA(0)->node_mem_map; |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 4368 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4369 | if (page_to_pfn(mem_map) != pgdat->node_start_pfn) |
Thomas Bogendoerfer | 467bc46 | 2008-01-08 15:33:11 -0800 | [diff] [blame] | 4370 | mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET); |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 4371 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4372 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4373 | #endif |
Andy Whitcroft | d41dee3 | 2005-06-23 00:07:54 -0700 | [diff] [blame] | 4374 | #endif /* CONFIG_FLAT_NODE_MEM_MAP */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4375 | } |
| 4376 | |
Johannes Weiner | 9109fb7 | 2008-07-23 21:27:20 -0700 | [diff] [blame] | 4377 | void __paginginit free_area_init_node(int nid, unsigned long *zones_size, |
| 4378 | unsigned long node_start_pfn, unsigned long *zholes_size) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4379 | { |
Johannes Weiner | 9109fb7 | 2008-07-23 21:27:20 -0700 | [diff] [blame] | 4380 | pg_data_t *pgdat = NODE_DATA(nid); |
| 4381 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4382 | pgdat->node_id = nid; |
| 4383 | pgdat->node_start_pfn = node_start_pfn; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4384 | calculate_node_totalpages(pgdat, zones_size, zholes_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4385 | |
| 4386 | alloc_node_mem_map(pgdat); |
Yinghai Lu | e8c27ac | 2008-06-01 13:15:22 -0700 | [diff] [blame] | 4387 | #ifdef CONFIG_FLAT_NODE_MEM_MAP |
| 4388 | printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n", |
| 4389 | nid, (unsigned long)pgdat, |
| 4390 | (unsigned long)pgdat->node_mem_map); |
| 4391 | #endif |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4392 | |
| 4393 | free_area_init_core(pgdat, zones_size, zholes_size); |
| 4394 | } |
| 4395 | |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 4396 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
Miklos Szeredi | 418508c | 2007-05-23 13:57:55 -0700 | [diff] [blame] | 4397 | |
| 4398 | #if MAX_NUMNODES > 1 |
| 4399 | /* |
| 4400 | * Figure out the number of possible node ids. |
| 4401 | */ |
| 4402 | static void __init setup_nr_node_ids(void) |
| 4403 | { |
| 4404 | unsigned int node; |
| 4405 | unsigned int highest = 0; |
| 4406 | |
| 4407 | for_each_node_mask(node, node_possible_map) |
| 4408 | highest = node; |
| 4409 | nr_node_ids = highest + 1; |
| 4410 | } |
| 4411 | #else |
| 4412 | static inline void setup_nr_node_ids(void) |
| 4413 | { |
| 4414 | } |
| 4415 | #endif |
| 4416 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4417 | /** |
Tejun Heo | 1e01979 | 2011-07-12 09:45:34 +0200 | [diff] [blame] | 4418 | * node_map_pfn_alignment - determine the maximum internode alignment |
| 4419 | * |
| 4420 | * This function should be called after node map is populated and sorted. |
| 4421 | * It calculates the maximum power of two alignment which can distinguish |
| 4422 | * all the nodes. |
| 4423 | * |
| 4424 | * For example, if all nodes are 1GiB and aligned to 1GiB, the return value |
| 4425 | * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the |
| 4426 | * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is |
| 4427 | * shifted, 1GiB is enough and this function will indicate so. |
| 4428 | * |
| 4429 | * This is used to test whether pfn -> nid mapping of the chosen memory |
| 4430 | * model has fine enough granularity to avoid incorrect mapping for the |
| 4431 | * populated node map. |
| 4432 | * |
| 4433 | * Returns the determined alignment in pfn's. 0 if there is no alignment |
| 4434 | * requirement (single node). |
| 4435 | */ |
| 4436 | unsigned long __init node_map_pfn_alignment(void) |
| 4437 | { |
| 4438 | unsigned long accl_mask = 0, last_end = 0; |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 4439 | unsigned long start, end, mask; |
Tejun Heo | 1e01979 | 2011-07-12 09:45:34 +0200 | [diff] [blame] | 4440 | int last_nid = -1; |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 4441 | int i, nid; |
Tejun Heo | 1e01979 | 2011-07-12 09:45:34 +0200 | [diff] [blame] | 4442 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 4443 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) { |
Tejun Heo | 1e01979 | 2011-07-12 09:45:34 +0200 | [diff] [blame] | 4444 | if (!start || last_nid < 0 || last_nid == nid) { |
| 4445 | last_nid = nid; |
| 4446 | last_end = end; |
| 4447 | continue; |
| 4448 | } |
| 4449 | |
| 4450 | /* |
| 4451 | * Start with a mask granular enough to pin-point to the |
| 4452 | * start pfn and tick off bits one-by-one until it becomes |
| 4453 | * too coarse to separate the current node from the last. |
| 4454 | */ |
| 4455 | mask = ~((1 << __ffs(start)) - 1); |
| 4456 | while (mask && last_end <= (start & (mask << 1))) |
| 4457 | mask <<= 1; |
| 4458 | |
| 4459 | /* accumulate all internode masks */ |
| 4460 | accl_mask |= mask; |
| 4461 | } |
| 4462 | |
| 4463 | /* convert mask to number of pages */ |
| 4464 | return ~accl_mask + 1; |
| 4465 | } |
| 4466 | |
Mel Gorman | a6af2bc | 2007-02-10 01:42:57 -0800 | [diff] [blame] | 4467 | /* Find the lowest pfn for a node */ |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 4468 | static unsigned long __init find_min_pfn_for_node(int nid) |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4469 | { |
Mel Gorman | a6af2bc | 2007-02-10 01:42:57 -0800 | [diff] [blame] | 4470 | unsigned long min_pfn = ULONG_MAX; |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 4471 | unsigned long start_pfn; |
| 4472 | int i; |
Mel Gorman | 1abbfb4 | 2006-11-23 12:01:41 +0000 | [diff] [blame] | 4473 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 4474 | for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL) |
| 4475 | min_pfn = min(min_pfn, start_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4476 | |
Mel Gorman | a6af2bc | 2007-02-10 01:42:57 -0800 | [diff] [blame] | 4477 | if (min_pfn == ULONG_MAX) { |
| 4478 | printk(KERN_WARNING |
Paul Jackson | 2bc0d26 | 2008-06-22 07:22:17 -0700 | [diff] [blame] | 4479 | "Could not find start_pfn for node %d\n", nid); |
Mel Gorman | a6af2bc | 2007-02-10 01:42:57 -0800 | [diff] [blame] | 4480 | return 0; |
| 4481 | } |
| 4482 | |
| 4483 | return min_pfn; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4484 | } |
| 4485 | |
| 4486 | /** |
| 4487 | * find_min_pfn_with_active_regions - Find the minimum PFN registered |
| 4488 | * |
| 4489 | * It returns the minimum PFN based on information provided via |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 4490 | * add_active_range(). |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4491 | */ |
| 4492 | unsigned long __init find_min_pfn_with_active_regions(void) |
| 4493 | { |
| 4494 | return find_min_pfn_for_node(MAX_NUMNODES); |
| 4495 | } |
| 4496 | |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 4497 | /* |
| 4498 | * early_calculate_totalpages() |
| 4499 | * Sum pages in active regions for movable zone. |
| 4500 | * Populate N_HIGH_MEMORY for calculating usable_nodes. |
| 4501 | */ |
Adrian Bunk | 484f51f | 2007-10-16 01:26:03 -0700 | [diff] [blame] | 4502 | static unsigned long __init early_calculate_totalpages(void) |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 4503 | { |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 4504 | unsigned long totalpages = 0; |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 4505 | unsigned long start_pfn, end_pfn; |
| 4506 | int i, nid; |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 4507 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 4508 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) { |
| 4509 | unsigned long pages = end_pfn - start_pfn; |
| 4510 | |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 4511 | totalpages += pages; |
| 4512 | if (pages) |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 4513 | node_set_state(nid, N_HIGH_MEMORY); |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 4514 | } |
| 4515 | return totalpages; |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 4516 | } |
| 4517 | |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4518 | /* |
| 4519 | * Find the PFN the Movable zone begins in each node. Kernel memory |
| 4520 | * is spread evenly between nodes as long as the nodes have enough |
| 4521 | * memory. When they don't, some nodes will have more kernelcore than |
| 4522 | * others |
| 4523 | */ |
Adrian Bunk | b69a728 | 2008-07-23 21:28:12 -0700 | [diff] [blame] | 4524 | static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn) |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4525 | { |
| 4526 | int i, nid; |
| 4527 | unsigned long usable_startpfn; |
| 4528 | unsigned long kernelcore_node, kernelcore_remaining; |
Yinghai Lu | 66918dc | 2009-06-30 11:41:37 -0700 | [diff] [blame] | 4529 | /* save the state before borrow the nodemask */ |
| 4530 | nodemask_t saved_node_state = node_states[N_HIGH_MEMORY]; |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 4531 | unsigned long totalpages = early_calculate_totalpages(); |
| 4532 | int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]); |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4533 | |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 4534 | /* |
| 4535 | * If movablecore was specified, calculate what size of |
| 4536 | * kernelcore that corresponds so that memory usable for |
| 4537 | * any allocation type is evenly spread. If both kernelcore |
| 4538 | * and movablecore are specified, then the value of kernelcore |
| 4539 | * will be used for required_kernelcore if it's greater than |
| 4540 | * what movablecore would have allowed. |
| 4541 | */ |
| 4542 | if (required_movablecore) { |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 4543 | unsigned long corepages; |
| 4544 | |
| 4545 | /* |
| 4546 | * Round-up so that ZONE_MOVABLE is at least as large as what |
| 4547 | * was requested by the user |
| 4548 | */ |
| 4549 | required_movablecore = |
| 4550 | roundup(required_movablecore, MAX_ORDER_NR_PAGES); |
| 4551 | corepages = totalpages - required_movablecore; |
| 4552 | |
| 4553 | required_kernelcore = max(required_kernelcore, corepages); |
| 4554 | } |
| 4555 | |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4556 | /* If kernelcore was not specified, there is no ZONE_MOVABLE */ |
| 4557 | if (!required_kernelcore) |
Yinghai Lu | 66918dc | 2009-06-30 11:41:37 -0700 | [diff] [blame] | 4558 | goto out; |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4559 | |
| 4560 | /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */ |
| 4561 | find_usable_zone_for_movable(); |
| 4562 | usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone]; |
| 4563 | |
| 4564 | restart: |
| 4565 | /* Spread kernelcore memory as evenly as possible throughout nodes */ |
| 4566 | kernelcore_node = required_kernelcore / usable_nodes; |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 4567 | for_each_node_state(nid, N_HIGH_MEMORY) { |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 4568 | unsigned long start_pfn, end_pfn; |
| 4569 | |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4570 | /* |
| 4571 | * Recalculate kernelcore_node if the division per node |
| 4572 | * now exceeds what is necessary to satisfy the requested |
| 4573 | * amount of memory for the kernel |
| 4574 | */ |
| 4575 | if (required_kernelcore < kernelcore_node) |
| 4576 | kernelcore_node = required_kernelcore / usable_nodes; |
| 4577 | |
| 4578 | /* |
| 4579 | * As the map is walked, we track how much memory is usable |
| 4580 | * by the kernel using kernelcore_remaining. When it is |
| 4581 | * 0, the rest of the node is usable by ZONE_MOVABLE |
| 4582 | */ |
| 4583 | kernelcore_remaining = kernelcore_node; |
| 4584 | |
| 4585 | /* Go through each range of PFNs within this node */ |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 4586 | for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) { |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4587 | unsigned long size_pages; |
| 4588 | |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 4589 | start_pfn = max(start_pfn, zone_movable_pfn[nid]); |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4590 | if (start_pfn >= end_pfn) |
| 4591 | continue; |
| 4592 | |
| 4593 | /* Account for what is only usable for kernelcore */ |
| 4594 | if (start_pfn < usable_startpfn) { |
| 4595 | unsigned long kernel_pages; |
| 4596 | kernel_pages = min(end_pfn, usable_startpfn) |
| 4597 | - start_pfn; |
| 4598 | |
| 4599 | kernelcore_remaining -= min(kernel_pages, |
| 4600 | kernelcore_remaining); |
| 4601 | required_kernelcore -= min(kernel_pages, |
| 4602 | required_kernelcore); |
| 4603 | |
| 4604 | /* Continue if range is now fully accounted */ |
| 4605 | if (end_pfn <= usable_startpfn) { |
| 4606 | |
| 4607 | /* |
| 4608 | * Push zone_movable_pfn to the end so |
| 4609 | * that if we have to rebalance |
| 4610 | * kernelcore across nodes, we will |
| 4611 | * not double account here |
| 4612 | */ |
| 4613 | zone_movable_pfn[nid] = end_pfn; |
| 4614 | continue; |
| 4615 | } |
| 4616 | start_pfn = usable_startpfn; |
| 4617 | } |
| 4618 | |
| 4619 | /* |
| 4620 | * The usable PFN range for ZONE_MOVABLE is from |
| 4621 | * start_pfn->end_pfn. Calculate size_pages as the |
| 4622 | * number of pages used as kernelcore |
| 4623 | */ |
| 4624 | size_pages = end_pfn - start_pfn; |
| 4625 | if (size_pages > kernelcore_remaining) |
| 4626 | size_pages = kernelcore_remaining; |
| 4627 | zone_movable_pfn[nid] = start_pfn + size_pages; |
| 4628 | |
| 4629 | /* |
| 4630 | * Some kernelcore has been met, update counts and |
| 4631 | * break if the kernelcore for this node has been |
| 4632 | * satisified |
| 4633 | */ |
| 4634 | required_kernelcore -= min(required_kernelcore, |
| 4635 | size_pages); |
| 4636 | kernelcore_remaining -= size_pages; |
| 4637 | if (!kernelcore_remaining) |
| 4638 | break; |
| 4639 | } |
| 4640 | } |
| 4641 | |
| 4642 | /* |
| 4643 | * If there is still required_kernelcore, we do another pass with one |
| 4644 | * less node in the count. This will push zone_movable_pfn[nid] further |
| 4645 | * along on the nodes that still have memory until kernelcore is |
| 4646 | * satisified |
| 4647 | */ |
| 4648 | usable_nodes--; |
| 4649 | if (usable_nodes && required_kernelcore > usable_nodes) |
| 4650 | goto restart; |
| 4651 | |
| 4652 | /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */ |
| 4653 | for (nid = 0; nid < MAX_NUMNODES; nid++) |
| 4654 | zone_movable_pfn[nid] = |
| 4655 | roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES); |
Yinghai Lu | 66918dc | 2009-06-30 11:41:37 -0700 | [diff] [blame] | 4656 | |
| 4657 | out: |
| 4658 | /* restore the node_state */ |
| 4659 | node_states[N_HIGH_MEMORY] = saved_node_state; |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4660 | } |
| 4661 | |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 4662 | /* Any regular memory on that node ? */ |
| 4663 | static void check_for_regular_memory(pg_data_t *pgdat) |
| 4664 | { |
| 4665 | #ifdef CONFIG_HIGHMEM |
| 4666 | enum zone_type zone_type; |
| 4667 | |
| 4668 | for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) { |
| 4669 | struct zone *zone = &pgdat->node_zones[zone_type]; |
Bob Liu | d0048b0 | 2012-01-12 17:19:07 -0800 | [diff] [blame] | 4670 | if (zone->present_pages) { |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 4671 | node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY); |
Bob Liu | d0048b0 | 2012-01-12 17:19:07 -0800 | [diff] [blame] | 4672 | break; |
| 4673 | } |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 4674 | } |
| 4675 | #endif |
| 4676 | } |
| 4677 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4678 | /** |
| 4679 | * free_area_init_nodes - Initialise all pg_data_t and zone data |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 4680 | * @max_zone_pfn: an array of max PFNs for each zone |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4681 | * |
| 4682 | * This will call free_area_init_node() for each active node in the system. |
| 4683 | * Using the page ranges provided by add_active_range(), the size of each |
| 4684 | * zone in each node and their holes is calculated. If the maximum PFN |
| 4685 | * between two adjacent zones match, it is assumed that the zone is empty. |
| 4686 | * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed |
| 4687 | * that arch_max_dma32_pfn has no pages. It is also assumed that a zone |
| 4688 | * starts where the previous one ended. For example, ZONE_DMA32 starts |
| 4689 | * at arch_max_dma_pfn. |
| 4690 | */ |
| 4691 | void __init free_area_init_nodes(unsigned long *max_zone_pfn) |
| 4692 | { |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 4693 | unsigned long start_pfn, end_pfn; |
| 4694 | int i, nid; |
Mel Gorman | a6af2bc | 2007-02-10 01:42:57 -0800 | [diff] [blame] | 4695 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4696 | /* Record where the zone boundaries are */ |
| 4697 | memset(arch_zone_lowest_possible_pfn, 0, |
| 4698 | sizeof(arch_zone_lowest_possible_pfn)); |
| 4699 | memset(arch_zone_highest_possible_pfn, 0, |
| 4700 | sizeof(arch_zone_highest_possible_pfn)); |
| 4701 | arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions(); |
| 4702 | arch_zone_highest_possible_pfn[0] = max_zone_pfn[0]; |
| 4703 | for (i = 1; i < MAX_NR_ZONES; i++) { |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4704 | if (i == ZONE_MOVABLE) |
| 4705 | continue; |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4706 | arch_zone_lowest_possible_pfn[i] = |
| 4707 | arch_zone_highest_possible_pfn[i-1]; |
| 4708 | arch_zone_highest_possible_pfn[i] = |
| 4709 | max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]); |
| 4710 | } |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4711 | arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0; |
| 4712 | arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0; |
| 4713 | |
| 4714 | /* Find the PFNs that ZONE_MOVABLE begins at in each node */ |
| 4715 | memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn)); |
| 4716 | find_zone_movable_pfns_for_nodes(zone_movable_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4717 | |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4718 | /* Print out the zone ranges */ |
| 4719 | printk("Zone PFN ranges:\n"); |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4720 | for (i = 0; i < MAX_NR_ZONES; i++) { |
| 4721 | if (i == ZONE_MOVABLE) |
| 4722 | continue; |
David Rientjes | 72f0ba0 | 2010-03-05 13:42:14 -0800 | [diff] [blame] | 4723 | printk(" %-8s ", zone_names[i]); |
| 4724 | if (arch_zone_lowest_possible_pfn[i] == |
| 4725 | arch_zone_highest_possible_pfn[i]) |
| 4726 | printk("empty\n"); |
| 4727 | else |
| 4728 | printk("%0#10lx -> %0#10lx\n", |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4729 | arch_zone_lowest_possible_pfn[i], |
| 4730 | arch_zone_highest_possible_pfn[i]); |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4731 | } |
| 4732 | |
| 4733 | /* Print out the PFNs ZONE_MOVABLE begins at in each node */ |
| 4734 | printk("Movable zone start PFN for each node\n"); |
| 4735 | for (i = 0; i < MAX_NUMNODES; i++) { |
| 4736 | if (zone_movable_pfn[i]) |
| 4737 | printk(" Node %d: %lu\n", i, zone_movable_pfn[i]); |
| 4738 | } |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4739 | |
| 4740 | /* Print out the early_node_map[] */ |
Tejun Heo | c13291a | 2011-07-12 10:46:30 +0200 | [diff] [blame] | 4741 | printk("Early memory PFN ranges\n"); |
| 4742 | for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) |
| 4743 | printk(" %3d: %0#10lx -> %0#10lx\n", nid, start_pfn, end_pfn); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4744 | |
| 4745 | /* Initialise every node */ |
Mel Gorman | 708614e | 2008-07-23 21:26:51 -0700 | [diff] [blame] | 4746 | mminit_verify_pageflags_layout(); |
Christoph Lameter | 8ef8286 | 2007-02-20 13:57:52 -0800 | [diff] [blame] | 4747 | setup_nr_node_ids(); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4748 | for_each_online_node(nid) { |
| 4749 | pg_data_t *pgdat = NODE_DATA(nid); |
Johannes Weiner | 9109fb7 | 2008-07-23 21:27:20 -0700 | [diff] [blame] | 4750 | free_area_init_node(nid, NULL, |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4751 | find_min_pfn_for_node(nid), NULL); |
Lee Schermerhorn | 37b07e4 | 2007-10-16 01:25:39 -0700 | [diff] [blame] | 4752 | |
| 4753 | /* Any memory on that node */ |
| 4754 | if (pgdat->node_present_pages) |
| 4755 | node_set_state(nid, N_HIGH_MEMORY); |
| 4756 | check_for_regular_memory(pgdat); |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4757 | } |
| 4758 | } |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4759 | |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 4760 | static int __init cmdline_parse_core(char *p, unsigned long *core) |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4761 | { |
| 4762 | unsigned long long coremem; |
| 4763 | if (!p) |
| 4764 | return -EINVAL; |
| 4765 | |
| 4766 | coremem = memparse(p, &p); |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 4767 | *core = coremem >> PAGE_SHIFT; |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4768 | |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 4769 | /* Paranoid check that UL is enough for the coremem value */ |
Mel Gorman | 2a1e274 | 2007-07-17 04:03:12 -0700 | [diff] [blame] | 4770 | WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX); |
| 4771 | |
| 4772 | return 0; |
| 4773 | } |
Mel Gorman | ed7ed36 | 2007-07-17 04:03:14 -0700 | [diff] [blame] | 4774 | |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 4775 | /* |
| 4776 | * kernelcore=size sets the amount of memory for use for allocations that |
| 4777 | * cannot be reclaimed or migrated. |
| 4778 | */ |
| 4779 | static int __init cmdline_parse_kernelcore(char *p) |
| 4780 | { |
| 4781 | return cmdline_parse_core(p, &required_kernelcore); |
| 4782 | } |
| 4783 | |
| 4784 | /* |
| 4785 | * movablecore=size sets the amount of memory for use for allocations that |
| 4786 | * can be reclaimed or migrated. |
| 4787 | */ |
| 4788 | static int __init cmdline_parse_movablecore(char *p) |
| 4789 | { |
| 4790 | return cmdline_parse_core(p, &required_movablecore); |
| 4791 | } |
| 4792 | |
Mel Gorman | ed7ed36 | 2007-07-17 04:03:14 -0700 | [diff] [blame] | 4793 | early_param("kernelcore", cmdline_parse_kernelcore); |
Mel Gorman | 7e63efe | 2007-07-17 04:03:15 -0700 | [diff] [blame] | 4794 | early_param("movablecore", cmdline_parse_movablecore); |
Mel Gorman | ed7ed36 | 2007-07-17 04:03:14 -0700 | [diff] [blame] | 4795 | |
Tejun Heo | 0ee332c | 2011-12-08 10:22:09 -0800 | [diff] [blame] | 4796 | #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */ |
Mel Gorman | c713216 | 2006-09-27 01:49:43 -0700 | [diff] [blame] | 4797 | |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 4798 | /** |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 4799 | * set_dma_reserve - set the specified number of pages reserved in the first zone |
| 4800 | * @new_dma_reserve: The number of pages to mark reserved |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 4801 | * |
| 4802 | * The per-cpu batchsize and zone watermarks are determined by present_pages. |
| 4803 | * In the DMA zone, a significant percentage may be consumed by kernel image |
| 4804 | * and other unfreeable allocations which can skew the watermarks badly. This |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 4805 | * function may optionally be used to account for unfreeable pages in the |
| 4806 | * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and |
| 4807 | * smaller per-cpu batchsize. |
Mel Gorman | 0e0b864 | 2006-09-27 01:49:56 -0700 | [diff] [blame] | 4808 | */ |
| 4809 | void __init set_dma_reserve(unsigned long new_dma_reserve) |
| 4810 | { |
| 4811 | dma_reserve = new_dma_reserve; |
| 4812 | } |
| 4813 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4814 | void __init free_area_init(unsigned long *zones_size) |
| 4815 | { |
Johannes Weiner | 9109fb7 | 2008-07-23 21:27:20 -0700 | [diff] [blame] | 4816 | free_area_init_node(0, zones_size, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4817 | __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL); |
| 4818 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4819 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4820 | static int page_alloc_cpu_notify(struct notifier_block *self, |
| 4821 | unsigned long action, void *hcpu) |
| 4822 | { |
| 4823 | int cpu = (unsigned long)hcpu; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4824 | |
Rafael J. Wysocki | 8bb7844 | 2007-05-09 02:35:10 -0700 | [diff] [blame] | 4825 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) { |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 4826 | drain_pages(cpu); |
| 4827 | |
| 4828 | /* |
| 4829 | * Spill the event counters of the dead processor |
| 4830 | * into the current processors event counters. |
| 4831 | * This artificially elevates the count of the current |
| 4832 | * processor. |
| 4833 | */ |
Christoph Lameter | f8891e5 | 2006-06-30 01:55:45 -0700 | [diff] [blame] | 4834 | vm_events_fold_cpu(cpu); |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 4835 | |
| 4836 | /* |
| 4837 | * Zero the differential counters of the dead processor |
| 4838 | * so that the vm statistics are consistent. |
| 4839 | * |
| 4840 | * This is only okay since the processor is dead and cannot |
| 4841 | * race with what we are doing. |
| 4842 | */ |
Christoph Lameter | 2244b95 | 2006-06-30 01:55:33 -0700 | [diff] [blame] | 4843 | refresh_cpu_vm_stats(cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4844 | } |
| 4845 | return NOTIFY_OK; |
| 4846 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4847 | |
| 4848 | void __init page_alloc_init(void) |
| 4849 | { |
| 4850 | hotcpu_notifier(page_alloc_cpu_notify, 0); |
| 4851 | } |
| 4852 | |
| 4853 | /* |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 4854 | * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio |
| 4855 | * or min_free_kbytes changes. |
| 4856 | */ |
| 4857 | static void calculate_totalreserve_pages(void) |
| 4858 | { |
| 4859 | struct pglist_data *pgdat; |
| 4860 | unsigned long reserve_pages = 0; |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 4861 | enum zone_type i, j; |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 4862 | |
| 4863 | for_each_online_pgdat(pgdat) { |
| 4864 | for (i = 0; i < MAX_NR_ZONES; i++) { |
| 4865 | struct zone *zone = pgdat->node_zones + i; |
| 4866 | unsigned long max = 0; |
| 4867 | |
| 4868 | /* Find valid and maximum lowmem_reserve in the zone */ |
| 4869 | for (j = i; j < MAX_NR_ZONES; j++) { |
| 4870 | if (zone->lowmem_reserve[j] > max) |
| 4871 | max = zone->lowmem_reserve[j]; |
| 4872 | } |
| 4873 | |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 4874 | /* we treat the high watermark as reserved pages. */ |
| 4875 | max += high_wmark_pages(zone); |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 4876 | |
| 4877 | if (max > zone->present_pages) |
| 4878 | max = zone->present_pages; |
| 4879 | reserve_pages += max; |
Johannes Weiner | ab8fabd | 2012-01-10 15:07:42 -0800 | [diff] [blame] | 4880 | /* |
| 4881 | * Lowmem reserves are not available to |
| 4882 | * GFP_HIGHUSER page cache allocations and |
| 4883 | * kswapd tries to balance zones to their high |
| 4884 | * watermark. As a result, neither should be |
| 4885 | * regarded as dirtyable memory, to prevent a |
| 4886 | * situation where reclaim has to clean pages |
| 4887 | * in order to balance the zones. |
| 4888 | */ |
| 4889 | zone->dirty_balance_reserve = max; |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 4890 | } |
| 4891 | } |
Johannes Weiner | ab8fabd | 2012-01-10 15:07:42 -0800 | [diff] [blame] | 4892 | dirty_balance_reserve = reserve_pages; |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 4893 | totalreserve_pages = reserve_pages; |
| 4894 | } |
| 4895 | |
| 4896 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4897 | * setup_per_zone_lowmem_reserve - called whenever |
| 4898 | * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone |
| 4899 | * has a correct pages reserved value, so an adequate number of |
| 4900 | * pages are left in the zone after a successful __alloc_pages(). |
| 4901 | */ |
| 4902 | static void setup_per_zone_lowmem_reserve(void) |
| 4903 | { |
| 4904 | struct pglist_data *pgdat; |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 4905 | enum zone_type j, idx; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4906 | |
KAMEZAWA Hiroyuki | ec936fc | 2006-03-27 01:15:59 -0800 | [diff] [blame] | 4907 | for_each_online_pgdat(pgdat) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4908 | for (j = 0; j < MAX_NR_ZONES; j++) { |
| 4909 | struct zone *zone = pgdat->node_zones + j; |
| 4910 | unsigned long present_pages = zone->present_pages; |
| 4911 | |
| 4912 | zone->lowmem_reserve[j] = 0; |
| 4913 | |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 4914 | idx = j; |
| 4915 | while (idx) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4916 | struct zone *lower_zone; |
| 4917 | |
Christoph Lameter | 2f6726e | 2006-09-25 23:31:18 -0700 | [diff] [blame] | 4918 | idx--; |
| 4919 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4920 | if (sysctl_lowmem_reserve_ratio[idx] < 1) |
| 4921 | sysctl_lowmem_reserve_ratio[idx] = 1; |
| 4922 | |
| 4923 | lower_zone = pgdat->node_zones + idx; |
| 4924 | lower_zone->lowmem_reserve[j] = present_pages / |
| 4925 | sysctl_lowmem_reserve_ratio[idx]; |
| 4926 | present_pages += lower_zone->present_pages; |
| 4927 | } |
| 4928 | } |
| 4929 | } |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 4930 | |
| 4931 | /* update totalreserve_pages */ |
| 4932 | calculate_totalreserve_pages(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4933 | } |
| 4934 | |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 4935 | /** |
Minchan Kim | bc75d33 | 2009-06-16 15:32:48 -0700 | [diff] [blame] | 4936 | * setup_per_zone_wmarks - called when min_free_kbytes changes |
Minchan Kim | bce7394 | 2009-06-16 15:32:50 -0700 | [diff] [blame] | 4937 | * or when memory is hot-{added|removed} |
Randy Dunlap | 88ca3b9 | 2006-10-04 02:15:25 -0700 | [diff] [blame] | 4938 | * |
Minchan Kim | bc75d33 | 2009-06-16 15:32:48 -0700 | [diff] [blame] | 4939 | * Ensures that the watermark[min,low,high] values for each zone are set |
| 4940 | * correctly with respect to min_free_kbytes. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4941 | */ |
Minchan Kim | bc75d33 | 2009-06-16 15:32:48 -0700 | [diff] [blame] | 4942 | void setup_per_zone_wmarks(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4943 | { |
| 4944 | unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); |
| 4945 | unsigned long lowmem_pages = 0; |
| 4946 | struct zone *zone; |
| 4947 | unsigned long flags; |
| 4948 | |
| 4949 | /* Calculate total number of !ZONE_HIGHMEM pages */ |
| 4950 | for_each_zone(zone) { |
| 4951 | if (!is_highmem(zone)) |
| 4952 | lowmem_pages += zone->present_pages; |
| 4953 | } |
| 4954 | |
| 4955 | for_each_zone(zone) { |
Andrew Morton | ac924c6 | 2006-05-15 09:43:59 -0700 | [diff] [blame] | 4956 | u64 tmp; |
| 4957 | |
Gerald Schaefer | 1125b4e | 2008-10-18 20:27:11 -0700 | [diff] [blame] | 4958 | spin_lock_irqsave(&zone->lock, flags); |
Andrew Morton | ac924c6 | 2006-05-15 09:43:59 -0700 | [diff] [blame] | 4959 | tmp = (u64)pages_min * zone->present_pages; |
| 4960 | do_div(tmp, lowmem_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4961 | if (is_highmem(zone)) { |
| 4962 | /* |
Nick Piggin | 669ed17 | 2005-11-13 16:06:45 -0800 | [diff] [blame] | 4963 | * __GFP_HIGH and PF_MEMALLOC allocations usually don't |
| 4964 | * need highmem pages, so cap pages_min to a small |
| 4965 | * value here. |
| 4966 | * |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 4967 | * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) |
Nick Piggin | 669ed17 | 2005-11-13 16:06:45 -0800 | [diff] [blame] | 4968 | * deltas controls asynch page reclaim, and so should |
| 4969 | * not be capped for highmem. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4970 | */ |
| 4971 | int min_pages; |
| 4972 | |
| 4973 | min_pages = zone->present_pages / 1024; |
| 4974 | if (min_pages < SWAP_CLUSTER_MAX) |
| 4975 | min_pages = SWAP_CLUSTER_MAX; |
| 4976 | if (min_pages > 128) |
| 4977 | min_pages = 128; |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 4978 | zone->watermark[WMARK_MIN] = min_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4979 | } else { |
Nick Piggin | 669ed17 | 2005-11-13 16:06:45 -0800 | [diff] [blame] | 4980 | /* |
| 4981 | * If it's a lowmem zone, reserve a number of pages |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4982 | * proportionate to the zone's size. |
| 4983 | */ |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 4984 | zone->watermark[WMARK_MIN] = tmp; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4985 | } |
| 4986 | |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 4987 | zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + (tmp >> 2); |
| 4988 | zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + (tmp >> 1); |
Mel Gorman | 56fd56b | 2007-10-16 01:25:58 -0700 | [diff] [blame] | 4989 | setup_zone_migrate_reserve(zone); |
Gerald Schaefer | 1125b4e | 2008-10-18 20:27:11 -0700 | [diff] [blame] | 4990 | spin_unlock_irqrestore(&zone->lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4991 | } |
Hideo AOKI | cb45b0e | 2006-04-10 22:52:59 -0700 | [diff] [blame] | 4992 | |
| 4993 | /* update totalreserve_pages */ |
| 4994 | calculate_totalreserve_pages(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4995 | } |
| 4996 | |
Randy Dunlap | 55a4462 | 2009-09-21 17:01:20 -0700 | [diff] [blame] | 4997 | /* |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 4998 | * The inactive anon list should be small enough that the VM never has to |
| 4999 | * do too much work, but large enough that each inactive page has a chance |
| 5000 | * to be referenced again before it is swapped out. |
| 5001 | * |
| 5002 | * The inactive_anon ratio is the target ratio of ACTIVE_ANON to |
| 5003 | * INACTIVE_ANON pages on this zone's LRU, maintained by the |
| 5004 | * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of |
| 5005 | * the anonymous pages are kept on the inactive list. |
| 5006 | * |
| 5007 | * total target max |
| 5008 | * memory ratio inactive anon |
| 5009 | * ------------------------------------- |
| 5010 | * 10MB 1 5MB |
| 5011 | * 100MB 1 50MB |
| 5012 | * 1GB 3 250MB |
| 5013 | * 10GB 10 0.9GB |
| 5014 | * 100GB 31 3GB |
| 5015 | * 1TB 101 10GB |
| 5016 | * 10TB 320 32GB |
| 5017 | */ |
KOSAKI Motohiro | 1b79acc | 2011-05-24 17:11:32 -0700 | [diff] [blame] | 5018 | static void __meminit calculate_zone_inactive_ratio(struct zone *zone) |
Minchan Kim | 96cb4df | 2009-06-16 15:32:49 -0700 | [diff] [blame] | 5019 | { |
| 5020 | unsigned int gb, ratio; |
| 5021 | |
| 5022 | /* Zone size in gigabytes */ |
| 5023 | gb = zone->present_pages >> (30 - PAGE_SHIFT); |
| 5024 | if (gb) |
| 5025 | ratio = int_sqrt(10 * gb); |
| 5026 | else |
| 5027 | ratio = 1; |
| 5028 | |
| 5029 | zone->inactive_ratio = ratio; |
| 5030 | } |
| 5031 | |
KOSAKI Motohiro | 839a4fc | 2011-05-24 17:11:31 -0700 | [diff] [blame] | 5032 | static void __meminit setup_per_zone_inactive_ratio(void) |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 5033 | { |
| 5034 | struct zone *zone; |
| 5035 | |
Minchan Kim | 96cb4df | 2009-06-16 15:32:49 -0700 | [diff] [blame] | 5036 | for_each_zone(zone) |
| 5037 | calculate_zone_inactive_ratio(zone); |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 5038 | } |
| 5039 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5040 | /* |
| 5041 | * Initialise min_free_kbytes. |
| 5042 | * |
| 5043 | * For small machines we want it small (128k min). For large machines |
| 5044 | * we want it large (64MB max). But it is not linear, because network |
| 5045 | * bandwidth does not increase linearly with machine size. We use |
| 5046 | * |
| 5047 | * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy: |
| 5048 | * min_free_kbytes = sqrt(lowmem_kbytes * 16) |
| 5049 | * |
| 5050 | * which yields |
| 5051 | * |
| 5052 | * 16MB: 512k |
| 5053 | * 32MB: 724k |
| 5054 | * 64MB: 1024k |
| 5055 | * 128MB: 1448k |
| 5056 | * 256MB: 2048k |
| 5057 | * 512MB: 2896k |
| 5058 | * 1024MB: 4096k |
| 5059 | * 2048MB: 5792k |
| 5060 | * 4096MB: 8192k |
| 5061 | * 8192MB: 11584k |
| 5062 | * 16384MB: 16384k |
| 5063 | */ |
KOSAKI Motohiro | 1b79acc | 2011-05-24 17:11:32 -0700 | [diff] [blame] | 5064 | int __meminit init_per_zone_wmark_min(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5065 | { |
| 5066 | unsigned long lowmem_kbytes; |
| 5067 | |
| 5068 | lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10); |
| 5069 | |
| 5070 | min_free_kbytes = int_sqrt(lowmem_kbytes * 16); |
| 5071 | if (min_free_kbytes < 128) |
| 5072 | min_free_kbytes = 128; |
| 5073 | if (min_free_kbytes > 65536) |
| 5074 | min_free_kbytes = 65536; |
Minchan Kim | bc75d33 | 2009-06-16 15:32:48 -0700 | [diff] [blame] | 5075 | setup_per_zone_wmarks(); |
KOSAKI Motohiro | a6cccdc | 2011-05-24 17:11:33 -0700 | [diff] [blame] | 5076 | refresh_zone_stat_thresholds(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5077 | setup_per_zone_lowmem_reserve(); |
Rik van Riel | 556adec | 2008-10-18 20:26:34 -0700 | [diff] [blame] | 5078 | setup_per_zone_inactive_ratio(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5079 | return 0; |
| 5080 | } |
Minchan Kim | bc75d33 | 2009-06-16 15:32:48 -0700 | [diff] [blame] | 5081 | module_init(init_per_zone_wmark_min) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5082 | |
| 5083 | /* |
| 5084 | * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so |
| 5085 | * that we can call two helper functions whenever min_free_kbytes |
| 5086 | * changes. |
| 5087 | */ |
| 5088 | int min_free_kbytes_sysctl_handler(ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 5089 | void __user *buffer, size_t *length, loff_t *ppos) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5090 | { |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 5091 | proc_dointvec(table, write, buffer, length, ppos); |
Mel Gorman | 3b1d92c | 2007-05-06 14:49:30 -0700 | [diff] [blame] | 5092 | if (write) |
Minchan Kim | bc75d33 | 2009-06-16 15:32:48 -0700 | [diff] [blame] | 5093 | setup_per_zone_wmarks(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5094 | return 0; |
| 5095 | } |
| 5096 | |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 5097 | #ifdef CONFIG_NUMA |
| 5098 | int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 5099 | void __user *buffer, size_t *length, loff_t *ppos) |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 5100 | { |
| 5101 | struct zone *zone; |
| 5102 | int rc; |
| 5103 | |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 5104 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 5105 | if (rc) |
| 5106 | return rc; |
| 5107 | |
| 5108 | for_each_zone(zone) |
Christoph Lameter | 8417bba | 2006-09-25 23:31:51 -0700 | [diff] [blame] | 5109 | zone->min_unmapped_pages = (zone->present_pages * |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 5110 | sysctl_min_unmapped_ratio) / 100; |
| 5111 | return 0; |
| 5112 | } |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 5113 | |
| 5114 | int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 5115 | void __user *buffer, size_t *length, loff_t *ppos) |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 5116 | { |
| 5117 | struct zone *zone; |
| 5118 | int rc; |
| 5119 | |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 5120 | rc = proc_dointvec_minmax(table, write, buffer, length, ppos); |
Christoph Lameter | 0ff3849 | 2006-09-25 23:31:52 -0700 | [diff] [blame] | 5121 | if (rc) |
| 5122 | return rc; |
| 5123 | |
| 5124 | for_each_zone(zone) |
| 5125 | zone->min_slab_pages = (zone->present_pages * |
| 5126 | sysctl_min_slab_ratio) / 100; |
| 5127 | return 0; |
| 5128 | } |
Christoph Lameter | 9614634 | 2006-07-03 00:24:13 -0700 | [diff] [blame] | 5129 | #endif |
| 5130 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5131 | /* |
| 5132 | * lowmem_reserve_ratio_sysctl_handler - just a wrapper around |
| 5133 | * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve() |
| 5134 | * whenever sysctl_lowmem_reserve_ratio changes. |
| 5135 | * |
| 5136 | * The reserve ratio obviously has absolutely no relation with the |
Mel Gorman | 4185896 | 2009-06-16 15:32:12 -0700 | [diff] [blame] | 5137 | * minimum watermarks. The lowmem reserve ratio can only make sense |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5138 | * if in function of the boot time zone sizes. |
| 5139 | */ |
| 5140 | int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 5141 | void __user *buffer, size_t *length, loff_t *ppos) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5142 | { |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 5143 | proc_dointvec_minmax(table, write, buffer, length, ppos); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5144 | setup_per_zone_lowmem_reserve(); |
| 5145 | return 0; |
| 5146 | } |
| 5147 | |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 5148 | /* |
| 5149 | * percpu_pagelist_fraction - changes the pcp->high for each zone on each |
| 5150 | * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist |
| 5151 | * can have before it gets flushed back to buddy allocator. |
| 5152 | */ |
| 5153 | |
| 5154 | int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 5155 | void __user *buffer, size_t *length, loff_t *ppos) |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 5156 | { |
| 5157 | struct zone *zone; |
| 5158 | unsigned int cpu; |
| 5159 | int ret; |
| 5160 | |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 5161 | ret = proc_dointvec_minmax(table, write, buffer, length, ppos); |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 5162 | if (!write || (ret == -EINVAL)) |
| 5163 | return ret; |
Dimitri Sivanich | 364df0e | 2009-06-23 12:37:04 -0700 | [diff] [blame] | 5164 | for_each_populated_zone(zone) { |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 5165 | for_each_possible_cpu(cpu) { |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 5166 | unsigned long high; |
| 5167 | high = zone->present_pages / percpu_pagelist_fraction; |
Christoph Lameter | 99dcc3e | 2010-01-05 15:34:51 +0900 | [diff] [blame] | 5168 | setup_pagelist_highmark( |
| 5169 | per_cpu_ptr(zone->pageset, cpu), high); |
Rohit Seth | 8ad4b1f | 2006-01-08 01:00:40 -0800 | [diff] [blame] | 5170 | } |
| 5171 | } |
| 5172 | return 0; |
| 5173 | } |
| 5174 | |
David S. Miller | f034b5d | 2006-08-24 03:08:07 -0700 | [diff] [blame] | 5175 | int hashdist = HASHDIST_DEFAULT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5176 | |
| 5177 | #ifdef CONFIG_NUMA |
| 5178 | static int __init set_hashdist(char *str) |
| 5179 | { |
| 5180 | if (!str) |
| 5181 | return 0; |
| 5182 | hashdist = simple_strtoul(str, &str, 0); |
| 5183 | return 1; |
| 5184 | } |
| 5185 | __setup("hashdist=", set_hashdist); |
| 5186 | #endif |
| 5187 | |
| 5188 | /* |
| 5189 | * allocate a large system hash table from bootmem |
| 5190 | * - it is assumed that the hash table must contain an exact power-of-2 |
| 5191 | * quantity of entries |
| 5192 | * - limit is the number of hash buckets, not the total allocation size |
| 5193 | */ |
| 5194 | void *__init alloc_large_system_hash(const char *tablename, |
| 5195 | unsigned long bucketsize, |
| 5196 | unsigned long numentries, |
| 5197 | int scale, |
| 5198 | int flags, |
| 5199 | unsigned int *_hash_shift, |
| 5200 | unsigned int *_hash_mask, |
| 5201 | unsigned long limit) |
| 5202 | { |
| 5203 | unsigned long long max = limit; |
| 5204 | unsigned long log2qty, size; |
| 5205 | void *table = NULL; |
| 5206 | |
| 5207 | /* allow the kernel cmdline to have a say */ |
| 5208 | if (!numentries) { |
| 5209 | /* round applicable memory size up to nearest megabyte */ |
Andrew Morton | 0490366 | 2006-12-06 20:37:33 -0800 | [diff] [blame] | 5210 | numentries = nr_kernel_pages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5211 | numentries += (1UL << (20 - PAGE_SHIFT)) - 1; |
| 5212 | numentries >>= 20 - PAGE_SHIFT; |
| 5213 | numentries <<= 20 - PAGE_SHIFT; |
| 5214 | |
| 5215 | /* limit to 1 bucket per 2^scale bytes of low memory */ |
| 5216 | if (scale > PAGE_SHIFT) |
| 5217 | numentries >>= (scale - PAGE_SHIFT); |
| 5218 | else |
| 5219 | numentries <<= (PAGE_SHIFT - scale); |
Paul Mundt | 9ab37b8 | 2007-01-05 16:36:30 -0800 | [diff] [blame] | 5220 | |
| 5221 | /* Make sure we've got at least a 0-order allocation.. */ |
Jan Beulich | 2c85f51 | 2009-09-21 17:03:07 -0700 | [diff] [blame] | 5222 | if (unlikely(flags & HASH_SMALL)) { |
| 5223 | /* Makes no sense without HASH_EARLY */ |
| 5224 | WARN_ON(!(flags & HASH_EARLY)); |
| 5225 | if (!(numentries >> *_hash_shift)) { |
| 5226 | numentries = 1UL << *_hash_shift; |
| 5227 | BUG_ON(!numentries); |
| 5228 | } |
| 5229 | } else if (unlikely((numentries * bucketsize) < PAGE_SIZE)) |
Paul Mundt | 9ab37b8 | 2007-01-05 16:36:30 -0800 | [diff] [blame] | 5230 | numentries = PAGE_SIZE / bucketsize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5231 | } |
John Hawkes | 6e692ed | 2006-03-25 03:08:02 -0800 | [diff] [blame] | 5232 | numentries = roundup_pow_of_two(numentries); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5233 | |
| 5234 | /* limit allocation size to 1/16 total memory by default */ |
| 5235 | if (max == 0) { |
| 5236 | max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4; |
| 5237 | do_div(max, bucketsize); |
| 5238 | } |
| 5239 | |
| 5240 | if (numentries > max) |
| 5241 | numentries = max; |
| 5242 | |
David Howells | f0d1b0b | 2006-12-08 02:37:49 -0800 | [diff] [blame] | 5243 | log2qty = ilog2(numentries); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5244 | |
| 5245 | do { |
| 5246 | size = bucketsize << log2qty; |
| 5247 | if (flags & HASH_EARLY) |
Jan Beulich | 74768ed | 2008-08-12 15:08:39 -0700 | [diff] [blame] | 5248 | table = alloc_bootmem_nopanic(size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5249 | else if (hashdist) |
| 5250 | table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL); |
| 5251 | else { |
Eric Dumazet | 1037b83 | 2007-07-15 23:38:05 -0700 | [diff] [blame] | 5252 | /* |
| 5253 | * If bucketsize is not a power-of-two, we may free |
Mel Gorman | a1dd268 | 2009-06-16 15:32:19 -0700 | [diff] [blame] | 5254 | * some pages at the end of hash table which |
| 5255 | * alloc_pages_exact() automatically does |
Eric Dumazet | 1037b83 | 2007-07-15 23:38:05 -0700 | [diff] [blame] | 5256 | */ |
Catalin Marinas | 264ef8a | 2009-07-07 10:33:01 +0100 | [diff] [blame] | 5257 | if (get_order(size) < MAX_ORDER) { |
Mel Gorman | a1dd268 | 2009-06-16 15:32:19 -0700 | [diff] [blame] | 5258 | table = alloc_pages_exact(size, GFP_ATOMIC); |
Catalin Marinas | 264ef8a | 2009-07-07 10:33:01 +0100 | [diff] [blame] | 5259 | kmemleak_alloc(table, size, 1, GFP_ATOMIC); |
| 5260 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5261 | } |
| 5262 | } while (!table && size > PAGE_SIZE && --log2qty); |
| 5263 | |
| 5264 | if (!table) |
| 5265 | panic("Failed to allocate %s hash table\n", tablename); |
| 5266 | |
Robin Holt | f241e660 | 2010-10-07 12:59:26 -0700 | [diff] [blame] | 5267 | printk(KERN_INFO "%s hash table entries: %ld (order: %d, %lu bytes)\n", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5268 | tablename, |
Robin Holt | f241e660 | 2010-10-07 12:59:26 -0700 | [diff] [blame] | 5269 | (1UL << log2qty), |
David Howells | f0d1b0b | 2006-12-08 02:37:49 -0800 | [diff] [blame] | 5270 | ilog2(size) - PAGE_SHIFT, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 5271 | size); |
| 5272 | |
| 5273 | if (_hash_shift) |
| 5274 | *_hash_shift = log2qty; |
| 5275 | if (_hash_mask) |
| 5276 | *_hash_mask = (1 << log2qty) - 1; |
| 5277 | |
| 5278 | return table; |
| 5279 | } |
KAMEZAWA Hiroyuki | a117e66 | 2006-03-27 01:15:25 -0800 | [diff] [blame] | 5280 | |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5281 | /* Return a pointer to the bitmap storing bits affecting a block of pages */ |
| 5282 | static inline unsigned long *get_pageblock_bitmap(struct zone *zone, |
| 5283 | unsigned long pfn) |
| 5284 | { |
| 5285 | #ifdef CONFIG_SPARSEMEM |
| 5286 | return __pfn_to_section(pfn)->pageblock_flags; |
| 5287 | #else |
| 5288 | return zone->pageblock_flags; |
| 5289 | #endif /* CONFIG_SPARSEMEM */ |
| 5290 | } |
Andrew Morton | 6220ec7 | 2006-10-19 23:29:05 -0700 | [diff] [blame] | 5291 | |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5292 | static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn) |
| 5293 | { |
| 5294 | #ifdef CONFIG_SPARSEMEM |
| 5295 | pfn &= (PAGES_PER_SECTION-1); |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5296 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5297 | #else |
| 5298 | pfn = pfn - zone->zone_start_pfn; |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5299 | return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS; |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5300 | #endif /* CONFIG_SPARSEMEM */ |
| 5301 | } |
| 5302 | |
| 5303 | /** |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5304 | * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5305 | * @page: The page within the block of interest |
| 5306 | * @start_bitidx: The first bit of interest to retrieve |
| 5307 | * @end_bitidx: The last bit of interest |
| 5308 | * returns pageblock_bits flags |
| 5309 | */ |
| 5310 | unsigned long get_pageblock_flags_group(struct page *page, |
| 5311 | int start_bitidx, int end_bitidx) |
| 5312 | { |
| 5313 | struct zone *zone; |
| 5314 | unsigned long *bitmap; |
| 5315 | unsigned long pfn, bitidx; |
| 5316 | unsigned long flags = 0; |
| 5317 | unsigned long value = 1; |
| 5318 | |
| 5319 | zone = page_zone(page); |
| 5320 | pfn = page_to_pfn(page); |
| 5321 | bitmap = get_pageblock_bitmap(zone, pfn); |
| 5322 | bitidx = pfn_to_bitidx(zone, pfn); |
| 5323 | |
| 5324 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) |
| 5325 | if (test_bit(bitidx + start_bitidx, bitmap)) |
| 5326 | flags |= value; |
| 5327 | |
| 5328 | return flags; |
| 5329 | } |
| 5330 | |
| 5331 | /** |
Mel Gorman | d9c2340 | 2007-10-16 01:26:01 -0700 | [diff] [blame] | 5332 | * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5333 | * @page: The page within the block of interest |
| 5334 | * @start_bitidx: The first bit of interest |
| 5335 | * @end_bitidx: The last bit of interest |
| 5336 | * @flags: The flags to set |
| 5337 | */ |
| 5338 | void set_pageblock_flags_group(struct page *page, unsigned long flags, |
| 5339 | int start_bitidx, int end_bitidx) |
| 5340 | { |
| 5341 | struct zone *zone; |
| 5342 | unsigned long *bitmap; |
| 5343 | unsigned long pfn, bitidx; |
| 5344 | unsigned long value = 1; |
| 5345 | |
| 5346 | zone = page_zone(page); |
| 5347 | pfn = page_to_pfn(page); |
| 5348 | bitmap = get_pageblock_bitmap(zone, pfn); |
| 5349 | bitidx = pfn_to_bitidx(zone, pfn); |
KAMEZAWA Hiroyuki | 86051ca | 2008-04-29 00:58:21 -0700 | [diff] [blame] | 5350 | VM_BUG_ON(pfn < zone->zone_start_pfn); |
| 5351 | VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages); |
Mel Gorman | 835c134 | 2007-10-16 01:25:47 -0700 | [diff] [blame] | 5352 | |
| 5353 | for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1) |
| 5354 | if (flags & value) |
| 5355 | __set_bit(bitidx + start_bitidx, bitmap); |
| 5356 | else |
| 5357 | __clear_bit(bitidx + start_bitidx, bitmap); |
| 5358 | } |
KAMEZAWA Hiroyuki | a5d76b5 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 5359 | |
| 5360 | /* |
| 5361 | * This is designed as sub function...plz see page_isolation.c also. |
| 5362 | * set/clear page block's type to be ISOLATE. |
| 5363 | * page allocater never alloc memory from ISOLATE block. |
| 5364 | */ |
| 5365 | |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 5366 | static int |
| 5367 | __count_immobile_pages(struct zone *zone, struct page *page, int count) |
| 5368 | { |
| 5369 | unsigned long pfn, iter, found; |
| 5370 | /* |
| 5371 | * For avoiding noise data, lru_add_drain_all() should be called |
| 5372 | * If ZONE_MOVABLE, the zone never contains immobile pages |
| 5373 | */ |
| 5374 | if (zone_idx(zone) == ZONE_MOVABLE) |
| 5375 | return true; |
| 5376 | |
| 5377 | if (get_pageblock_migratetype(page) == MIGRATE_MOVABLE) |
| 5378 | return true; |
| 5379 | |
| 5380 | pfn = page_to_pfn(page); |
| 5381 | for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) { |
| 5382 | unsigned long check = pfn + iter; |
| 5383 | |
Namhyung Kim | 29723fc | 2011-02-25 14:44:25 -0800 | [diff] [blame] | 5384 | if (!pfn_valid_within(check)) |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 5385 | continue; |
Namhyung Kim | 29723fc | 2011-02-25 14:44:25 -0800 | [diff] [blame] | 5386 | |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 5387 | page = pfn_to_page(check); |
| 5388 | if (!page_count(page)) { |
| 5389 | if (PageBuddy(page)) |
| 5390 | iter += (1 << page_order(page)) - 1; |
| 5391 | continue; |
| 5392 | } |
| 5393 | if (!PageLRU(page)) |
| 5394 | found++; |
| 5395 | /* |
| 5396 | * If there are RECLAIMABLE pages, we need to check it. |
| 5397 | * But now, memory offline itself doesn't call shrink_slab() |
| 5398 | * and it still to be fixed. |
| 5399 | */ |
| 5400 | /* |
| 5401 | * If the page is not RAM, page_count()should be 0. |
| 5402 | * we don't need more check. This is an _used_ not-movable page. |
| 5403 | * |
| 5404 | * The problematic thing here is PG_reserved pages. PG_reserved |
| 5405 | * is set to both of a memory hole page and a _used_ kernel |
| 5406 | * page at boot. |
| 5407 | */ |
| 5408 | if (found > count) |
| 5409 | return false; |
| 5410 | } |
| 5411 | return true; |
| 5412 | } |
| 5413 | |
| 5414 | bool is_pageblock_removable_nolock(struct page *page) |
| 5415 | { |
| 5416 | struct zone *zone = page_zone(page); |
| 5417 | return __count_immobile_pages(zone, page, 0); |
| 5418 | } |
| 5419 | |
KAMEZAWA Hiroyuki | a5d76b5 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 5420 | int set_migratetype_isolate(struct page *page) |
| 5421 | { |
| 5422 | struct zone *zone; |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 5423 | unsigned long flags, pfn; |
Robert Jennings | 925cc71 | 2009-12-17 14:44:38 +0000 | [diff] [blame] | 5424 | struct memory_isolate_notify arg; |
| 5425 | int notifier_ret; |
KAMEZAWA Hiroyuki | a5d76b5 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 5426 | int ret = -EBUSY; |
| 5427 | |
| 5428 | zone = page_zone(page); |
Robert Jennings | 925cc71 | 2009-12-17 14:44:38 +0000 | [diff] [blame] | 5429 | |
KAMEZAWA Hiroyuki | a5d76b5 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 5430 | spin_lock_irqsave(&zone->lock, flags); |
Robert Jennings | 925cc71 | 2009-12-17 14:44:38 +0000 | [diff] [blame] | 5431 | |
| 5432 | pfn = page_to_pfn(page); |
| 5433 | arg.start_pfn = pfn; |
| 5434 | arg.nr_pages = pageblock_nr_pages; |
| 5435 | arg.pages_found = 0; |
| 5436 | |
| 5437 | /* |
| 5438 | * It may be possible to isolate a pageblock even if the |
| 5439 | * migratetype is not MIGRATE_MOVABLE. The memory isolation |
| 5440 | * notifier chain is used by balloon drivers to return the |
| 5441 | * number of pages in a range that are held by the balloon |
| 5442 | * driver to shrink memory. If all the pages are accounted for |
| 5443 | * by balloons, are free, or on the LRU, isolation can continue. |
| 5444 | * Later, for example, when memory hotplug notifier runs, these |
| 5445 | * pages reported as "can be isolated" should be isolated(freed) |
| 5446 | * by the balloon driver through the memory notifier chain. |
| 5447 | */ |
| 5448 | notifier_ret = memory_isolate_notify(MEM_ISOLATE_COUNT, &arg); |
| 5449 | notifier_ret = notifier_to_errno(notifier_ret); |
KAMEZAWA Hiroyuki | 4b20477 | 2010-10-26 14:21:29 -0700 | [diff] [blame] | 5450 | if (notifier_ret) |
Robert Jennings | 925cc71 | 2009-12-17 14:44:38 +0000 | [diff] [blame] | 5451 | goto out; |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 5452 | /* |
| 5453 | * FIXME: Now, memory hotplug doesn't call shrink_slab() by itself. |
| 5454 | * We just check MOVABLE pages. |
| 5455 | */ |
| 5456 | if (__count_immobile_pages(zone, page, arg.pages_found)) |
Robert Jennings | 925cc71 | 2009-12-17 14:44:38 +0000 | [diff] [blame] | 5457 | ret = 0; |
| 5458 | |
KAMEZAWA Hiroyuki | 49ac825 | 2010-10-26 14:21:30 -0700 | [diff] [blame] | 5459 | /* |
| 5460 | * immobile means "not-on-lru" paes. If immobile is larger than |
| 5461 | * removable-by-driver pages reported by notifier, we'll fail. |
| 5462 | */ |
| 5463 | |
KAMEZAWA Hiroyuki | a5d76b5 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 5464 | out: |
Robert Jennings | 925cc71 | 2009-12-17 14:44:38 +0000 | [diff] [blame] | 5465 | if (!ret) { |
| 5466 | set_pageblock_migratetype(page, MIGRATE_ISOLATE); |
| 5467 | move_freepages_block(zone, page, MIGRATE_ISOLATE); |
| 5468 | } |
| 5469 | |
KAMEZAWA Hiroyuki | a5d76b5 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 5470 | spin_unlock_irqrestore(&zone->lock, flags); |
| 5471 | if (!ret) |
Christoph Lameter | 9f8f217 | 2008-02-04 22:29:11 -0800 | [diff] [blame] | 5472 | drain_all_pages(); |
KAMEZAWA Hiroyuki | a5d76b5 | 2007-10-16 01:26:11 -0700 | [diff] [blame] | 5473 | return ret; |
| 5474 | } |
| 5475 | |
| 5476 | void unset_migratetype_isolate(struct page *page) |
| 5477 | { |
| 5478 | struct zone *zone; |
| 5479 | unsigned long flags; |
| 5480 | zone = page_zone(page); |
| 5481 | spin_lock_irqsave(&zone->lock, flags); |
| 5482 | if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE) |
| 5483 | goto out; |
| 5484 | set_pageblock_migratetype(page, MIGRATE_MOVABLE); |
| 5485 | move_freepages_block(zone, page, MIGRATE_MOVABLE); |
| 5486 | out: |
| 5487 | spin_unlock_irqrestore(&zone->lock, flags); |
| 5488 | } |
KAMEZAWA Hiroyuki | 0c0e619 | 2007-10-16 01:26:12 -0700 | [diff] [blame] | 5489 | |
| 5490 | #ifdef CONFIG_MEMORY_HOTREMOVE |
| 5491 | /* |
| 5492 | * All pages in the range must be isolated before calling this. |
| 5493 | */ |
| 5494 | void |
| 5495 | __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn) |
| 5496 | { |
| 5497 | struct page *page; |
| 5498 | struct zone *zone; |
| 5499 | int order, i; |
| 5500 | unsigned long pfn; |
| 5501 | unsigned long flags; |
| 5502 | /* find the first valid pfn */ |
| 5503 | for (pfn = start_pfn; pfn < end_pfn; pfn++) |
| 5504 | if (pfn_valid(pfn)) |
| 5505 | break; |
| 5506 | if (pfn == end_pfn) |
| 5507 | return; |
| 5508 | zone = page_zone(pfn_to_page(pfn)); |
| 5509 | spin_lock_irqsave(&zone->lock, flags); |
| 5510 | pfn = start_pfn; |
| 5511 | while (pfn < end_pfn) { |
| 5512 | if (!pfn_valid(pfn)) { |
| 5513 | pfn++; |
| 5514 | continue; |
| 5515 | } |
| 5516 | page = pfn_to_page(pfn); |
| 5517 | BUG_ON(page_count(page)); |
| 5518 | BUG_ON(!PageBuddy(page)); |
| 5519 | order = page_order(page); |
| 5520 | #ifdef CONFIG_DEBUG_VM |
| 5521 | printk(KERN_INFO "remove from free list %lx %d %lx\n", |
| 5522 | pfn, 1 << order, end_pfn); |
| 5523 | #endif |
| 5524 | list_del(&page->lru); |
| 5525 | rmv_page_order(page); |
| 5526 | zone->free_area[order].nr_free--; |
| 5527 | __mod_zone_page_state(zone, NR_FREE_PAGES, |
| 5528 | - (1UL << order)); |
| 5529 | for (i = 0; i < (1 << order); i++) |
| 5530 | SetPageReserved((page+i)); |
| 5531 | pfn += (1 << order); |
| 5532 | } |
| 5533 | spin_unlock_irqrestore(&zone->lock, flags); |
| 5534 | } |
| 5535 | #endif |
Wu Fengguang | 8d22ba1 | 2009-12-16 12:19:58 +0100 | [diff] [blame] | 5536 | |
| 5537 | #ifdef CONFIG_MEMORY_FAILURE |
| 5538 | bool is_free_buddy_page(struct page *page) |
| 5539 | { |
| 5540 | struct zone *zone = page_zone(page); |
| 5541 | unsigned long pfn = page_to_pfn(page); |
| 5542 | unsigned long flags; |
| 5543 | int order; |
| 5544 | |
| 5545 | spin_lock_irqsave(&zone->lock, flags); |
| 5546 | for (order = 0; order < MAX_ORDER; order++) { |
| 5547 | struct page *page_head = page - (pfn & ((1 << order) - 1)); |
| 5548 | |
| 5549 | if (PageBuddy(page_head) && page_order(page_head) >= order) |
| 5550 | break; |
| 5551 | } |
| 5552 | spin_unlock_irqrestore(&zone->lock, flags); |
| 5553 | |
| 5554 | return order < MAX_ORDER; |
| 5555 | } |
| 5556 | #endif |
Wu Fengguang | 718a382 | 2010-03-10 15:20:43 -0800 | [diff] [blame] | 5557 | |
| 5558 | static struct trace_print_flags pageflag_names[] = { |
| 5559 | {1UL << PG_locked, "locked" }, |
| 5560 | {1UL << PG_error, "error" }, |
| 5561 | {1UL << PG_referenced, "referenced" }, |
| 5562 | {1UL << PG_uptodate, "uptodate" }, |
| 5563 | {1UL << PG_dirty, "dirty" }, |
| 5564 | {1UL << PG_lru, "lru" }, |
| 5565 | {1UL << PG_active, "active" }, |
| 5566 | {1UL << PG_slab, "slab" }, |
| 5567 | {1UL << PG_owner_priv_1, "owner_priv_1" }, |
| 5568 | {1UL << PG_arch_1, "arch_1" }, |
| 5569 | {1UL << PG_reserved, "reserved" }, |
| 5570 | {1UL << PG_private, "private" }, |
| 5571 | {1UL << PG_private_2, "private_2" }, |
| 5572 | {1UL << PG_writeback, "writeback" }, |
| 5573 | #ifdef CONFIG_PAGEFLAGS_EXTENDED |
| 5574 | {1UL << PG_head, "head" }, |
| 5575 | {1UL << PG_tail, "tail" }, |
| 5576 | #else |
| 5577 | {1UL << PG_compound, "compound" }, |
| 5578 | #endif |
| 5579 | {1UL << PG_swapcache, "swapcache" }, |
| 5580 | {1UL << PG_mappedtodisk, "mappedtodisk" }, |
| 5581 | {1UL << PG_reclaim, "reclaim" }, |
Wu Fengguang | 718a382 | 2010-03-10 15:20:43 -0800 | [diff] [blame] | 5582 | {1UL << PG_swapbacked, "swapbacked" }, |
| 5583 | {1UL << PG_unevictable, "unevictable" }, |
| 5584 | #ifdef CONFIG_MMU |
| 5585 | {1UL << PG_mlocked, "mlocked" }, |
| 5586 | #endif |
| 5587 | #ifdef CONFIG_ARCH_USES_PG_UNCACHED |
| 5588 | {1UL << PG_uncached, "uncached" }, |
| 5589 | #endif |
| 5590 | #ifdef CONFIG_MEMORY_FAILURE |
| 5591 | {1UL << PG_hwpoison, "hwpoison" }, |
| 5592 | #endif |
| 5593 | {-1UL, NULL }, |
| 5594 | }; |
| 5595 | |
| 5596 | static void dump_page_flags(unsigned long flags) |
| 5597 | { |
| 5598 | const char *delim = ""; |
| 5599 | unsigned long mask; |
| 5600 | int i; |
| 5601 | |
| 5602 | printk(KERN_ALERT "page flags: %#lx(", flags); |
| 5603 | |
| 5604 | /* remove zone id */ |
| 5605 | flags &= (1UL << NR_PAGEFLAGS) - 1; |
| 5606 | |
| 5607 | for (i = 0; pageflag_names[i].name && flags; i++) { |
| 5608 | |
| 5609 | mask = pageflag_names[i].mask; |
| 5610 | if ((flags & mask) != mask) |
| 5611 | continue; |
| 5612 | |
| 5613 | flags &= ~mask; |
| 5614 | printk("%s%s", delim, pageflag_names[i].name); |
| 5615 | delim = "|"; |
| 5616 | } |
| 5617 | |
| 5618 | /* check for left over flags */ |
| 5619 | if (flags) |
| 5620 | printk("%s%#lx", delim, flags); |
| 5621 | |
| 5622 | printk(")\n"); |
| 5623 | } |
| 5624 | |
| 5625 | void dump_page(struct page *page) |
| 5626 | { |
| 5627 | printk(KERN_ALERT |
| 5628 | "page:%p count:%d mapcount:%d mapping:%p index:%#lx\n", |
Andrea Arcangeli | 4e9f64c | 2011-01-13 15:46:29 -0800 | [diff] [blame] | 5629 | page, atomic_read(&page->_count), page_mapcount(page), |
Wu Fengguang | 718a382 | 2010-03-10 15:20:43 -0800 | [diff] [blame] | 5630 | page->mapping, page->index); |
| 5631 | dump_page_flags(page->flags); |
Daisuke Nishimura | f212ad7 | 2011-03-23 16:42:25 -0700 | [diff] [blame] | 5632 | mem_cgroup_print_bad_page(page); |
Wu Fengguang | 718a382 | 2010-03-10 15:20:43 -0800 | [diff] [blame] | 5633 | } |